Context-driven learning of human-object interactions

Abstract

A method for classifying human-object interactions includes identifying human-object interactions in an input. Contextual features of the input are identified. Each identified contextual feature is compared to the identified human-object interaction. An importance of the identified contextual feature to the identified human-object interaction is determined. The contextual feature is fused with the identified human-object interaction when the importance is greater than a threshold.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to Greek Patent Application No. 20190100515, filed on November 15, 2019, entitled “Context-driven learning of human-object interactions”, the disclosure of which is expressly incorporated herein by reference in its entirety.

[0003] Public domain

[0004] Aspects of the present disclosure relate generally to object classification, and more particularly to object classification based on context and human-object appearance.

[0005] background

[0006] An artificial neural network, which may include a group of interconnected artificial neurons (e.g., a neuron model), may refer to a computing device or a method to be performed by a computing device. The artificial neural network may be used in various applications and / or devices, such as Internet Protocol (IP) cameras, Internet of Things (IoT) devices, autonomous vehicles, and / or service robots.

[0007] Context can be used to classify visual scenes. For example, context can determine objects or actions in an image, co-occurrence of objects, and / or spatial locations of objects. In some cases, context can be used to improve human-object interaction classification. In most cases, conventional classification systems always consider context when classifying human-object interactions. It is desirable to improve the use of context in classification systems.

[0008] Overview

[0009] In one aspect of the present disclosure, a method for classifying an input is provided. The method includes identifying a human-object interaction in the input. The method also includes identifying contextual features of the input. Additionally, the method includes comparing each identified contextual feature with the identified human-object interaction. The method also includes determining the importance of the identified contextual feature to the identified human-object interaction. In addition, the method includes fusing the contextual feature with the identified human-object interaction when the importance is greater than a threshold.

[0010] In another aspect of the present disclosure, a device for classifying input is provided. The device includes a memory and one or more processors coupled to the memory. The processor(s) are configured to identify human-object interactions in the input. The processor(s) are also configured to identify contextual features of the input. In addition, the processor(s) are configured to compare each identified contextual feature with the identified human-object interaction. The processor(s) are also configured to determine the importance of the identified contextual feature for the identified human-object interaction. In addition, the processor(s) are configured to fuse the contextual feature with the identified human-object interaction when the importance is greater than a threshold.

[0011] In another aspect of the present disclosure, a device for classifying an input is provided. The device includes means for identifying a human-object interaction in the input. The device also includes means for identifying contextual features of the input. Additionally, the device includes means for comparing each identified contextual feature with the identified human-object interaction. The device also includes means for determining the importance of the identified contextual feature to the identified human-object interaction. In addition, the device includes means for fusing the contextual feature with the identified human-object interaction when the importance is greater than a threshold.

[0012] According to another aspect of the present disclosure, a non-transient computer-readable medium is provided. The computer-readable medium is encoded with a program code for classifying an input. The program code is executed by a processor and includes a code for identifying a human-object interaction in the input. The program code also includes a code for identifying a contextual feature of the input. Additionally, the program code includes a code for comparing each identified contextual feature with an identified human-object interaction. The program code also includes a code for determining the importance of the identified contextual feature to the identified human-object interaction. In addition, the program code includes a code for fusing the contextual feature with the identified human-object interaction when the importance is greater than a threshold.

[0013] Additional features and advantages of the present disclosure will be described below. Those skilled in the art will appreciate that the present disclosure can be easily used as a basis for modifying or designing other structures for implementing the same purpose as the present disclosure. Those skilled in the art will also recognize that such equivalent constructions do not depart from the teachings of the present disclosure as set forth in the appended claims. The novel features that are considered to be characteristic of the present disclosure, both in terms of its organization and method of operation, together with further objects and advantages, will be better understood when the following description is considered in conjunction with the accompanying drawings. However, it is to be clearly understood that each of the drawings is provided for illustration and description purposes only and is not intended as a definition of limitations of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The features, nature and advantages of the present disclosure will become more apparent when the detailed description set forth below is read in conjunction with the accompanying drawings, in which like reference numerals designate correspondingly throughout.

[0016] Figure 1 An example implementation of designing a neural network using a system on a chip (SOC) including a general-purpose processor in accordance with certain aspects of the present disclosure is illustrated.

[0017] Figure 2A , 2B 2C are diagrams illustrating neural networks according to aspects of the present disclosure.

[0018] Figure 2D is a diagram illustrating an exemplary deep convolutional network (DCN) according to aspects of the present disclosure.

[0019] Figure 3 is a block diagram illustrating an exemplary deep convolutional network (DCN) according to aspects of the present disclosure.

[0020] Figure 4A An example of a general classification system is explained.

[0021] Figure 4B An example of an adaptive fusion neural network in accordance with aspects of the present disclosure is illustrated.

[0022] Figure 5A An example of a forward pass operation of an adaptive model according to aspects of the present disclosure is illustrated.

[0023] Figure 5B An example of a reverse pass operation according to aspects of the present disclosure is illustrated.

[0024] Figure 6 An overview of an adaptive model according to aspects of the present disclosure is illustrated.

[0025] Figure 7 A flow chart illustrating a method for classifying input in accordance with aspects of the present disclosure is illustrated.

[0026] Detailed Description

[0027] The detailed description set forth below in conjunction with the accompanying drawings is intended as a description of various configurations and is not intended to represent the only configurations in which the described concepts may be practiced. This detailed description includes specific details in order to provide a thorough understanding of the various concepts. However, it will be apparent to those skilled in the art that these concepts may be practiced without these specific details. In some instances, well-known structures and components are shown in block diagram form to avoid obscuring such concepts.

[0028] Based on this teaching, it will be appreciated by those skilled in the art that the scope of the present disclosure is intended to cover any aspect of the present disclosure, whether it is implemented independently or in combination with any other aspect of the present disclosure. For example, any number of aspects described can be used to implement a device or practice method. In addition, the scope of the present disclosure is intended to cover such devices or methods that are practiced using other structures, functionality, or structures and functionality that are supplemented or different from the various aspects of the present disclosure described. It should be understood that any aspect of the present disclosure disclosed can be implemented by one or more elements of the claims.

[0029] The word “exemplary” is used herein to mean “serving as an example, instance, or illustration.” Any aspect described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other aspects.

[0030] Although specific aspects are described, numerous variations and permutations of these aspects fall within the scope of the present disclosure. Although some benefits and advantages of preferred aspects are mentioned, the scope of the present disclosure is not intended to be limited to specific benefits, uses or goals. On the contrary, various aspects of the present disclosure are intended to be broadly applicable to different technologies, system configurations, networks and protocols, some of which are illustrated as examples in the accompanying drawings and the following description of preferred aspects. The detailed description and drawings merely illustrate the present disclosure and do not limit the present disclosure, and the scope of the present disclosure is defined by the attached claims and their equivalent technical solutions.

[0031] Human-object interactions can be characterized by a relationship (verb, noun) pair (such as (ride, bicycle) or (walk, horse)) that describes the relationship between the human and the object of the interaction. In some cases, context clues are used to improve the classification of human-object interactions. Context can be obtained from the surrounding environment (e.g., scene). The surrounding environment can identify possible sets of interactions. Additionally or alternatively, context can also be obtained from co-occurring objects. For example, a person may be surrounded by books when reading in a library, or may be surrounded by other bicycles when competing in a cycling championship. Some systems also incorporate body part context in the classification model.

[0032] Conventional techniques for combining human-object appearance and context include early fusion or late fusion. In a late fusion system, human appearance and simultaneous object appearance are encoded separately by non-weight-sharing neural networks. The predictions of the non-weight-sharing neural networks are combined in the form of late fusion at the classifier layer. In an early fusion system, the neural network combines the context representation and the human-object representation at an early stage. The neural network processes the combination jointly to generate an interactive prediction. In the present disclosure, the context representation may also be referred to as the context representation.

[0033] Human-object interactions provide a rich set of contextual information that traditional systems may not be able to exploit. For example, early fusion systems are not scalable because the combination of multiple contextual features increases the feature dimension, which leads to system performance degradation. The increase in feature dimension may lead to limiting the use of context to the global surrounding environment or object co-occurrence. In addition, late fusion systems do not exploit the correlation between human-object and contextual resources. Therefore, late fusion systems cannot apply contextual information adaptively.

[0034] Aspects of the present disclosure relate to a system that adaptively uses contextual information when classifying human-object interactions. In one configuration, a set of contextual information is defined to improve interaction recognition (e.g., classification). Additionally, a gating mechanism learns to identify when contextual information should be used to classify human-object interactions.

[0035] Figure 1 An example implementation of a system on a chip (SOC) 100 according to certain aspects of the present disclosure is illustrated, which may include a central processing unit (CPU) 102 or a multi-core CPU configured for adaptive context representation and human-object representation fusion via a neural network. Variables (e.g., neural signals and synaptic weights), system parameters associated with a computing device (e.g., a neural network with weights), delays, frequency bin information, and task information may be stored in a memory block associated with a neural processing unit (NPU) 108, a memory block associated with the CPU 102, a memory block associated with a graphics processing unit (GPU) 104, a memory block associated with a digital signal processor (DSP) 106, a memory block 118, or may be distributed across multiple blocks. Instructions executed at the CPU 102 may be loaded from a program memory associated with the CPU 102 or may be loaded from the memory block 118.

[0036] The SOC 100 may also include additional processing blocks customized for specific functions, such as a GPU 104, a DSP 106, a connectivity block 110 (which may include fifth generation (5G) connectivity, fourth generation long term evolution (4G LTE) connectivity, Wi-Fi connectivity, USB connectivity, Bluetooth connectivity, etc.), and a multimedia processor 112 that may detect and recognize gestures, for example. In one implementation, the NPU is implemented in the CPU, DSP, and / or GPU. The SOC 100 may also include a sensor processor 114, an image signal processor (ISP) 116, and / or a navigation module 120 (which may include a global positioning system).

[0037] SOC 100 may be based on the ARM instruction set. In one aspect of the present disclosure, the instructions loaded into the general purpose processor 102 may include code for identifying human-object interactions in the input. The general purpose processor 102 may also include code for identifying contextual features of the input. The general purpose processor 102 may further include code for comparing each identified contextual feature with the identified human-object interaction. The general purpose processor 102 may also include code for determining the importance of the identified contextual feature for the identified human-object interaction. Finally, the general purpose processor 102 may include code for fusing the contextual feature with the identified human-object interaction when the importance is greater than a threshold.

[0038] Deep learning architectures can perform object recognition tasks by learning to represent the input at successively higher levels of abstraction in each layer, thereby building useful feature representations of the input data. In this way, deep learning addresses a major bottleneck of traditional machine learning. Before the advent of deep learning, machine learning approaches to object recognition problems might rely heavily on human-engineered features, perhaps combined with shallow classifiers. A shallow classifier could be a two-class linear classifier, for example, where a weighted sum of the components of a feature vector is compared to a threshold to predict which class the input belongs to. A human-engineered feature could be a template or kernel that is customized for a specific problem domain by an engineer with domain expertise. In contrast, a deep learning architecture can learn to represent features similar to what a human engineer might design, but it does so through training. In addition, deep networks can learn to represent and recognize new types of features that humans might not have considered.

[0039] A deep learning architecture can learn a hierarchy of features. For example, if a first layer is presented with visual data, the first layer can learn to recognize relatively simple features (such as edges) in the input stream. In another example, if a first layer is presented with auditory data, the first layer can learn to recognize spectral power in specific frequencies. A second layer that takes the output of the first layer as input can learn to recognize combinations of features, such as simple shapes for visual data or combinations of sounds for auditory data. For example, higher layers can learn to represent complex shapes in visual data or words in auditory data. Even higher layers can learn to recognize common visual objects or spoken phrases.

[0040] Deep learning architectures can perform particularly well when applied to problems that have a natural hierarchical structure. For example, the classification of motor vehicles can benefit from first learning to recognize wheels, windshields, and other features. These features can be combined in different ways at higher levels to recognize cars, trucks, and airplanes.

[0041] Neural networks can be designed to have various connectivity patterns. In a feedforward network, information is passed from lower layers to higher layers, with each neuron in a given layer communicating to neurons in a higher layer. As described above, hierarchical representations can be constructed in successive layers of a feedforward network. Neural networks can also have reflow or feedback (also known as top-down) connections. In a reflow connection, the output from a neuron in a given layer can be communicated to another neuron in the same layer. The reflow architecture can help identify patterns that span more than one block of input data delivered sequentially to the neural network. The connection from a neuron in a given layer to a neuron in a lower layer is called a feedback (or top-down) connection. A network with many feedback connections may be helpful when the recognition of high-level concepts can assist in discerning specific low-level features of the input.

[0042] The connections between layers of a neural network can be fully connected or partially connected. Figure 2A An example of a fully connected neural network 202 is illustrated. In the fully connected neural network 202, a neuron in a first layer may communicate its output to each neuron in a second layer, so that each neuron in the second layer will receive input from each neuron in the first layer. Figure 2B An example of a locally connected neural network 204 is illustrated. In the locally connected neural network 204, neurons in a first layer may be connected to a limited number of neurons in a second layer. More generally, the locally connected layers of the locally connected neural network 204 may be configured so that each neuron in a layer will have the same or similar connectivity pattern, but their connection strengths may have different values ​​(e.g., 210, 212, 214, and 216). The locally connected connectivity patterns may produce spatially distinct receptive fields in higher layers because higher layer neurons in a given area may receive inputs that are tuned through training to the properties of a limited portion of the total input to the network.

[0043] An example of a locally connected neural network is a convolutional neural network. Figure 2C An example of a convolutional neural network 206 is illustrated. The convolutional neural network 206 may be configured such that the connection strengths associated with the inputs to each neuron in the second layer are shared (e.g., 208). Convolutional neural networks may be well suited for problems where the spatial location of the inputs is meaningful.

[0044] One type of convolutional neural network is a deep convolutional network (DCN). Figure 2DA detailed example of a DCN 200 designed to recognize visual features from an image 226 input from an image capture device 230 (such as an onboard camera) is illustrated. The DCN 200 of the current example can be trained to identify traffic signs and numbers provided on traffic signs. Of course, the DCN 200 can be trained for other tasks, such as identifying lane markings or identifying traffic lights.

[0045] The DCN 200 may be trained using supervised learning. During training, an image (such as an image 226 of a speed limit sign) may be presented to the DCN 200, and a "forward pass" may then be calculated to produce an output 222. The DCN 200 may include a feature extraction section and a classification section. Upon receiving the image 226, a convolution layer 232 may apply a convolution kernel (not shown) to the image 226 to generate a first set of feature maps 218. As an example, the convolution kernel of the convolution layer 232 may be a 5x5 kernel that generates a 28x28 feature map. In this example, since four different feature maps are generated in the first set of feature maps 218, four different convolution kernels are applied to the image 226 at the convolution layer 232. A convolution kernel may also be referred to as a filter or a convolution filter.

[0046] The first set of feature maps 218 may be subsampled by a max pooling layer (not shown) to generate a second set of feature maps 220. The max pooling layer reduces the size of the first set of feature maps 218. That is, the size of the second set of feature maps 220 (such as 14x14) is smaller than the size of the first set of feature maps 218 (such as 28x28). The reduced size provides similar information to subsequent layers while reducing memory consumption. The second set of feature maps 220 may be further convolved via one or more subsequent convolutional layers (not shown) to generate a subsequent set of one or more sets of feature maps (not shown).

[0047] exist Figure 2D In the example of , the second set of feature maps 220 is convolved to generate a first feature vector 224. In addition, the first feature vector 224 is further convolved to generate a second feature vector 228. Each feature of the second feature vector 228 may include a number corresponding to a possible feature of the image 226 (such as "sign", "60", and "100"). A softmax function (not shown) can convert the numbers in the second feature vector 228 into probabilities. In this way, the output 222 of the DCN 200 is the probability that the image 226 includes one or more features.

[0048] In this example, the probabilities for "logo" and "60" in output 222 are higher than the probabilities for other features of output 222 (such as "30", "40", "50", "70", "80", "90", and "100"). Before training, output 222 generated by DCN 200 is likely to be incorrect. Thus, the error between output 222 and the target output can be calculated. The target output is the true value of image 226 (e.g., "logo" and "60"). The weights of DCN 200 can then be adjusted to align the output 222 of DCN 200 more closely with the target output.

[0049] To adjust the weights, the learning algorithm may calculate a gradient vector for the weights. The gradient may indicate the amount by which the error will increase or decrease if the weights are adjusted. At the top layer, the gradient may correspond directly to the value of the weights connecting the activated neurons in the penultimate layer to the neurons in the output layer. In lower layers, the gradient may depend on the value of the weights and the calculated error gradient of the higher layers. The weights may then be adjusted to reduce the error. This way of adjusting weights may be referred to as "backward propagation" because it involves a "backward pass" in a neural network.

[0050] In practice, the error gradient of the weights may be calculated over a small number of examples so that the calculated gradient approximates the true error gradient. This approximation method may be referred to as stochastic gradient descent. Stochastic gradient descent may be repeated until the error rate achievable by the entire system has stopped decreasing or until the error rate has reached a target level. After learning, the DCN may be presented with a new image (e.g., image 226 of a speed limit sign) and a forward pass through the network may produce output 222, which may be considered an inference or prediction of the DCN.

[0051] Deep belief network (DBN) is a probabilistic model including multiple layers of hidden nodes. DBN can be used to extract hierarchical representations of training data sets. DBN can be obtained by stacking multiple layers of restricted Boltzmann machines (RBM). RBM is a class of artificial neural networks that can learn probability distributions on input sets. Since RBM can learn probability distributions without information about which class each input should be classified into, RBM is often used in unsupervised learning. Using a hybrid unsupervised and supervised paradigm, the bottom RBM of DBN can be trained in an unsupervised manner and can be used as a feature extractor, while the top RBM can be trained in a supervised manner (on the joint distribution of the input and target class from the previous layer) and can be used as a classifier.

[0052] A deep convolutional network (DCN) is a network of convolutional networks configured with additional pooling and normalization layers. DCN has achieved state-of-the-art performance on many tasks. DCN can be trained using supervised learning, where both the input and output targets are known for many examples and are used to modify the weights of the network by using gradient descent.

[0053] The DCN can be a feed-forward network. In addition, as described above, the connections from the neurons in the first layer of the DCN to the neuron groups in the next higher layer are shared across the neurons in the first layer. The feed-forward and shared connections of the DCN can be used for fast processing. The computational burden of the DCN can be much smaller than, for example, the computational burden of a neural network of similar size that includes recurrent or feedback connections.

[0054] The processing of each layer of the convolutional network can be considered as a spatially invariant template or basis projection. If the input is first decomposed into multiple channels, such as the red, green and blue channels of a color image, the convolutional network trained on the input can be considered to be three-dimensional, with two spatial dimensions along the axis of the image and a third dimension that captures color information. The output of the convolutional connection can be considered to form a feature map in a subsequent layer, each element in the feature map (e.g., 220) receives input from a certain range of neurons in the previous layer (e.g., feature map 218) and from each channel in the multiple channels. The values ​​in the feature map can be further processed with nonlinearity (such as correction, max (0, x)). The values ​​from adjacent neurons can be further pooled (which corresponds to downsampling) and can provide additional local invariance and dimensionality reduction. Normalization can also be applied by lateral inhibition between neurons in the feature map, which corresponds to whitening.

[0055] The performance of deep learning architectures can improve as more labeled data points become available or as computing power increases. Modern deep neural networks are routinely trained with thousands of times more computing resources than were available to a typical researcher just fifteen years ago. New architectures and training paradigms can further boost the performance of deep learning. Rectified linear units can reduce the training problem known as vanishing gradients. New training techniques can reduce over-fitting and thus enable larger models to achieve better generalization. Encapsulation techniques can abstract the data within a given receptive field and further improve overall performance.

[0056] Figure 3 is a block diagram illustrating a deep convolutional network 350. The deep convolutional network 350 may include multiple layers of different types based on connectivity and weight sharing. Figure 3As shown in FIG. 3 , the deep convolution network 350 includes convolution blocks 354A and 354B. Each of the convolution blocks 354A and 354B may be configured with a convolution layer (CONV) 356 , a normalization layer (LNorm) 358 , and a maximum pooling layer (MAX POOL) 360 .

[0057] The convolution layer 356 may include one or more convolution filters that can be applied to the input data to generate a feature map. Although only two convolution blocks 354A, 354B are shown, the present disclosure is not limited thereto, but instead any number of convolution blocks 354A, 354B may be included in the deep convolution network 350 according to design preferences. The normalization layer 358 may normalize the output of the convolution filter. For example, the normalization layer 358 may provide whitening or lateral suppression. The maximum pooling layer 360 may provide downsampling aggregation in space to achieve local invariance and dimensionality reduction.

[0058] For example, the parallel filter banks of the deep convolutional network may be loaded onto the CPU 102 or GPU 104 of the SOC 100 to achieve high performance and low power consumption. In alternative embodiments, the parallel filter banks may be loaded onto the DSP 106 or ISP 116 of the SOC 100. In addition, the deep convolutional network 350 may access other processing blocks that may be present on the SOC 100, such as the sensor processor 114 and the navigation module 120 dedicated to sensors and navigation, respectively.

[0059] The deep convolutional network 350 may also include one or more fully connected layers 362 (FC1 and FC2). The deep convolutional network 350 may further include a logistic regression (LR) layer 364. Between each layer 356, 358, 360, 362, 364 of the deep convolutional network 350 are weights (not shown) to be updated. The output of each layer (e.g., 356, 358, 360, 362, 364) can be used as an input to a subsequent layer (e.g., 356, 358, 360, 362, 364) in the deep convolutional network 350 to learn a hierarchical feature representation from the input data 352 (e.g., image, audio, video, sensor data, and / or other input data) supplied from the first convolutional block 354A. The output of the deep convolutional network 350 is a classification score 366 for the input data 352. The classification score 366 can be a set of probabilities, where each probability is a probability that the input data includes a feature from a feature set.

[0060] Figure 4A An example of a general classification system is explained. Figure 4A In the example of , image 400 is provided to neural networks 410, 412 for classification. Image 400 includes person 402, object 404 (eg, horse), and context 406 (eg, equestrian center).

[0061] As discussed, the early fusion neural network 410 combines the context representation 420 (e.g., context features) and the human-object representation 422 at an early stage (e.g., before classification). The early fusion neural network 410 processes the combination to generate an interaction prediction.

[0062] If also Figure 4A As shown in FIG, the late fusion neural network 412 separately encodes the human-object representation 422 and the context representation 420. The late fusion neural network 412 may be a non-weight-sharing neural network and is used to separately encode the representations 422, 420. The predictions of the non-weight-sharing neural networks are combined at the classifier layer (not shown).

[0063] Figure 4B An example of an adaptive fusion neural network 450 according to aspects of the present disclosure is illustrated. Figure 4B As shown in , the input from the context representation 420 to the adaptive fusion neural network 450 is gated. Figure 4A The adaptive fusion neural network 450 may generate class probabilities for pairs of (verbs, nouns) describing relationships between humans and one or more objects in the image.

[0064] In one configuration, an adaptive fusion neural network generates a human-object feature vector f{action}. The human-object features may be obtained from the penultimate layer of a convolutional neural network (CNN). The adaptive fusion neural network receives an image I and human detection b{h} and object detection b{o} boxes that locate humans and objects in the image I. The adaptive fusion neural network generates a fixed-size feature vector Rd summarizing the appearance of humans and objects for interaction prediction. The feature vector Rd may be coupled with contextual features.

[0065] The adaptive fusion neural network also generates a context feature vector f{context}. The context features can be obtained from the penultimate layer of the CNN. In one configuration, multiple sources of context information are combined. For simplicity, in the current example, the context feature vector f{context} is now a fixed-size feature vector Rm containing multiple context features. As discussed, the context features can be adaptively combined with human-object features.

[0066] The final feature representation f{final} is generated based on an optional concatenation of a human-object feature vector f{action} and a context feature vector f{context}, such that f{final} = [f{action}, λ(f{context})]. The parameter λ is a free binary parameter that determines whether the context will be used (one if used, zero if not). The final feature representation f{final} is input to an interaction classifier f{classifier}, which outputs the probability of a (verb, noun) pair. In one configuration, during training, a standard cross entropy loss is determined based on the probabilities. Each subnetwork can be updated based on the loss.

[0067] The context submodule determines whether contextual features are helpful in classifying an interaction in an image. For example, scene context from a horse race will be relevant for classifying an interaction (riding, horse) but will not be relevant for an interaction (eating, sandwich). The relevance of the context may be related to the humans and objects that appear in the environment.

[0068] In one configuration, a context gating submodule f{gate} gates the context. The input to the context gating submodule f{gate} is the concatenation of the action and context [f{action}, f{context}] of dimension (d+m). The output of f{gate} is λ. As discussed, λ provides the importance of the context. A binary (e.g., discrete) output of 1 or 0 is used. The binary output specifies a non-differentiable operation.

[0069] In one configuration, a sigmoid nonlinearity is applied to the output of the context gating submodule (e.g., sigmoid(f{gate})). During training, the neural network may degenerate to a specific value, such as 1 or 0. To avoid a degenerate solution (e.g., always outputting 1 or 0), random noise can be extracted from a Gaussian distribution N (m=0, s=1). A hard thresholding operation can be applied to sigmoid(f{gate}). The hard thresholding operation converts the output to 1 (above the threshold) or 0 (below the threshold).

[0070] Non-differentiable operations cannot be used to obtain gradients and therefore cannot be trained. For this reason, straight-through estimators can be used to bypass non-differentiability. Straight-through estimators provide asymmetry for forward and backward propagation in neural networks. In the forward pass, for inference, a threshold operation is applied and predictions are based on the threshold. In the backward pass, the threshold operation is bypassed and the gradients are propagated through the continuous values ​​of the gate outputs.

[0071] Figure 5A An example of a forward pass operation 500 of an adaptive model 502 according to aspects of the present disclosure is illustrated. Figure 5AAs shown in , context representation 520 and human-object representation 522 can be input to the neural network. The fully connected layer 504 of the neural network generates a non-linear mapping 506, which operates on the context representation 520. The non-linear mapping 506 normalizes the dimensions of different context representations 520 so that the context representation 520 and the human-object representation 522 can be fused / gated together. The context gating submodule 508 generates an output based on the cascade of action and context [f{action}, f{context}]. The sigmoid function 510 is applied to the output of the context gating submodule 508, and such output of the sigmoid function 510 is a value between 0 and 1.

[0072] A hard threshold operation 512 may be applied to the output of the sigmoid function 510. If the output is greater than the threshold (e.g., greater than 0.5), the output is set to 1. If the output is less than the threshold (e.g., less than 0.5), the output is set to 0. The threshold (0.5) is for illustrative purposes only. In this example, a value of 1 gates the context representation 520 to the human-object representation 522. A value of 0 does not gate the context representation 520 to the human-object representation 522.

[0073] Figure 5B An example of a back propagation operation 550 of an adaptive model 502 according to aspects of the present disclosure is illustrated. Figure 5B As shown in , threshold operation 512 is bypassed 552 (indicated by an X on the connection), and the gradient is propagated through the continuous value of the gate output.

[0074] As discussed, the framework of the adaptive model 502 can incorporate one or more contextual representations. The contextual representation can include, for example, one or more of clothing, human-object spatial relationships, scene location, scene geometry, and / or co-occurring objects.

[0075] Feature activations from a pre-trained CNN can be used to extract clothing representations. To model objects that occur simultaneously with the interaction, object regions can be sampled from the area surrounding the human. Category predictions can be obtained from an object detection network. A one-hot encoded vector is generated to represent the category of objects present near the human.

[0076] Human-object spatial relations can be encoded by initializing an initial feature map with the same size as the image. One map encodes the location of the human, while the other encodes the location of the object.

[0077] The location of the scene provides clues about the forms of interactions that may be performed in the scene. For example, a sky area may indicate a kite flying interaction (fly, kite), while a road would indicate a car driving interaction (drive, car) or a bicycle riding interaction (ride, bicycle). To this end, semantic fragments of the scene can be extracted and activations can be used as descriptors of the location of the scene.

[0078] Figure 6 An overview of an adaptive model 600 according to aspects of the present disclosure is illustrated. Figure 6 As shown in FIG. 6 , the adaptive model 600 includes a feature extraction module 604 that receives an input image 602. The feature extraction module 604 may be a convolutional neural network (e.g., Figure 3 The feature extraction module 604 determines context representations Z1 to ZM based on various context representations available in the scene (e.g., clothing, human-object spatial relationships, scene location, scene geometry, and / or co-occurring objects). The context representations Z1 to ZM are determined based on the features fC1 to fCM processed by the convolutional neural network. Figure 6 , I is an input image 602 to a human-object feature extractor. The P blocks (e.g., PM, P1, PHO) are fully connected layers of the neural network. The P blocks independently process the M contextual features extracted from the image I to provide modality-specific preprocessing such as image normalization, object segmentation (for object context), and / or scene segmentation (for local scene context). Because the same image can be used for human-object or context extraction, the feature extraction module 604 can also determine a human-object representation X based on humans and objects in the scene depicted in the image I. The human-object representation X is determined based on the features fHO processed by the convolutional neural network.

[0079] For each context representation Z1 to ZM, the classification model 606 determines whether the context representation should be gated with the human-object representation X to determine the probability of a (verb, noun) pair. Figure 6 As shown in , the classification model 606 includes a gating module 608 and a classifier 610.

[0080] For each context representation Z1 to ZM, the gating module 608 receives a human-object representation X vector of size N x C and a context representation Z vector of size M x C'. A first gating function hΘ converts the human-object representation X vector to a second human-object representation Q of size N x C'. Functions hΘ and gΦ map the human-object representation and representation to the same subspace, respectively, so that they can be compared for compatibility via dot products. The compatibility score modulates the input context features using gΨ.

[0081] The context representation Z generates the second and third context vectors V, K of size M x C'. The dot product of the second human-object representation Q and the third context vector K generates a one-dimensional similarity scalar score α for each context feature Z of size M x N. The similarity scalar score α determines whether the context representation is related to the human-object representation. For example, the library location has nothing to do with the interaction between a person and a sandwich. On the contrary, the library location can be related to the interaction between a person and a book.

[0082] The dot product of the similarity scalar score α and the second context vector V generates a gated output vector Z' of size 1 x C'. That is, the similarity scalar score α can be used to reweight the context representation Z for further calculations (eg, classifying inputs using the modulated context representation via classifier 610).

[0083] Figure 7 1 illustrates a flow chart of a method 700 for classifying input according to aspects of the present disclosure. Figure 7 As shown in FIG, at block 702, a neural network identifies human-object interactions in an input. The input may be images of humans and objects in an environment. At block 704, the neural network identifies contextual features of the input. The contextual features may help classify human-object interactions. The contextual features may include clothing, human-object spatial relationships, scene locations, scene geometry, and / or objects occurring simultaneously.

[0084] At block 706, the neural network compares each identified contextual feature to the identified human-object interaction. At block 708, the neural network determines the importance of the identified contextual feature to the identified human-object interaction. At block 710, the neural network fuses the contextual feature with the identified human-object interaction when the importance is greater than a threshold.

[0085] In one configuration, a neural network determines the probability of a verb and noun pair based on a fused representation of the human-object interaction and context. The probability can be used to classify an input human-object interaction.

[0086] In an optional configuration, during training, noise is added to the context representation to strengthen the importance predictions in the training forward pass. In one configuration, the noise is removed during the training backward pass. The classifier can be robust to horizontal and vertical flips. Additionally, in an optional configuration, the positions of the human bounding boxes and / or object bounding boxes are jittered during training. The jittering provides robustness to jitter in the input.

[0087] In some aspects, method 700 may be performed by SOC 100 ( Figure 1That is, as an example and not limitation, each element of method 700 may be performed by SOC 100 or one or more processors (eg, CPU 102) and / or other included components.

[0088] Implementation examples are described in the following numbered clauses:

[0089] 1. A method for classifying an input, comprising:

[0090] identifying human-object interactions in the input;

[0091] Identify contextual features of the input;

[0092] comparing each identified context feature to the identified human-object interaction;

[0093] determining the importance of the identified contextual features to the identified human-object interaction; and

[0094] The context feature is fused with the identified human-object interaction when the importance is greater than a threshold.

[0095] 2. The method of clause 1, wherein the contextual features include one or more of clothing, human body part features, human-object spatial relationships, scene location, scene geometry, or co-occurring objects.

[0096] 3. The method of clause 1, further comprising determining the probability of verb and noun pairs based on the fused representation of human-object interaction and contextual features.

[0097] 4. The method of clause 1, further comprising adding noise during a forward pass of training to enhance importance determination.

[0098] 5. The method of clause 4, further comprising removing noise during the training back pass.

[0099] 6. A method as in clause 1, wherein the classifier used to classify the input is robust to horizontal and vertical flipping.

[0100] 7. The method of any of clauses 1-6, further comprising jittering the positions of human bounding boxes and / or object bounding boxes during training.

[0101] 8. An apparatus comprising:

[0102] Memory; and

[0103] at least one processor coupled to the memory, the at least one processor configured to:

[0104] Identify human-object interactions in input;

[0105] Identify contextual features of the input;

[0106] comparing each identified context feature to the identified human-object interaction;

[0107] determining a significance of the identified contextual feature to the identified human-object interaction; and

[0108] The context feature is fused with the identified human-object interaction when the importance is greater than a threshold.

[0109] 9. An apparatus as in clause 8, wherein the contextual features include clothing, human-object spatial relationships, scene location, scene geometry, and / or co-occurring objects.

[0110] 10. The apparatus of clause 8, wherein the at least one processor is further configured to determine a probability of a verb and noun pair based on a fused representation of the human-object interaction and contextual features.

[0111] 11. The apparatus of clause 8, wherein the at least one processor is further configured to add noise during a training forward pass to enhance importance determination.

[0112] 12. The apparatus of clause 11, wherein the at least one processor is further configured to remove noise during a training back pass.

[0113] 13. An apparatus as in clause 8, wherein the classifier used to classify the input is robust to horizontal and vertical flipping.

[0114] 14. The apparatus of any of clauses 8-13, wherein the at least one processor is further configured to jitter the positions of the human bounding boxes and / or the object bounding boxes during training.

[0115] 15. A device comprising:

[0116] means for identifying human-object interactions in input;

[0117] means for identifying contextual features of the input;

[0118] means for comparing each identified contextual feature to the identified human-object interaction;

[0119] means for determining the importance of the identified contextual feature to the identified human-object interaction; and

[0120] Means for fusing the contextual feature with the identified human-object interaction when the significance is greater than a threshold.

[0121] 16. The apparatus of clause 15, wherein the contextual features include one or more of clothing, human body part features, human-object spatial relationships, scene location, scene geometry, or co-occurring objects.

[0122] 17. The apparatus of clause 15, further comprising means for determining a probability of a verb and noun pair based on the fused representation of the human-object interaction and the contextual features.

[0123] 18. The apparatus of clause 15, further comprising means for adding noise during the training forward pass to enhance the importance determination.

[0124] 19. The apparatus of clause 18, further comprising means for removing noise during the training back-propagation.

[0125] 20. The apparatus of clause 15, wherein the classifier used to classify the input is robust to horizontal and vertical flipping.

[0126] 21. The apparatus of any of clauses 15-20, further comprising means for jittering the positions of human bounding boxes and / or object bounding boxes during training.

[0127] 22. A non-transitory computer readable medium having program code recorded thereon, the program code being executed by a processor and comprising:

[0128] program code for identifying human-object interactions in input;

[0129] program code for identifying contextual features of the input;

[0130] program code for comparing each identified contextual feature to the identified human-object interaction;

[0131] program code for determining the importance of the identified contextual features to the identified human-object interaction; and

[0132] Program code is provided for fusing the contextual feature with the identified human-object interaction when the significance is greater than a threshold.

[0133] 23. The non-transitory computer-readable medium of clause 22, wherein the contextual features include one or more of clothing, human-object spatial relationships, scene location, scene geometry, or co-occurring objects.

[0134] 24. The non-transitory computer-readable medium of clause 22, further comprising program code for determining a probability of a verb and noun pair based on the fused representation of the human-object interaction and the contextual features.

[0135] 25. The non-transitory computer readable medium of clause 22, further comprising program code for adding noise during a training forward pass to enhance importance determination.

[0136] 26. The non-transitory computer readable medium of clause 25, further comprising program code for removing noise during a training back pass.

[0137] 27. The non-transitory computer-readable medium of clause 22, wherein the classifier used to classify the input is robust to horizontal and vertical flipping.

[0138] 28. The non-transitory computer-readable medium of any of clauses 22-27, further comprising program code for jittering the positions of human bounding boxes and / or object bounding boxes during training.

[0139] The various operations of the methods described above may be performed by any suitable device capable of performing the corresponding functions. These devices may include various hardware and / or software components and / or modules, including but not limited to circuits, application specific integrated circuits (ASICs), or processors. In general, where there are operations illustrated in the accompanying drawings, those operations may have corresponding paired device-plus-function components with similar numbers.

[0140] As used, the term "determining" encompasses a wide variety of actions. For example, "determining" may include calculating, computing, processing, deriving, investigating, searching (e.g., searching in a table, a database, or another data structure), ascertaining, and the like. Additionally, "determining" may include receiving (e.g., receiving information), accessing (e.g., accessing data in a memory), and the like. Furthermore, "determining" may include resolving, selecting, choosing, establishing, and the like.

[0141] As used, a phrase referring to "at least one of" a list of items refers to any combination of those items, including single members. As an example, "at least one of a, b, or c" is intended to cover: a, b, c, ab, ac, bc, and abc.

[0142] The various illustrative logical blocks, modules, and circuits described in conjunction with the present disclosure may be implemented or performed with a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array signal (FPGA) or other programmable logic device (PLD), discrete gate or transistor logic, discrete hardware components, or any combination thereof, designed to perform the functions described. A general purpose processor may be a microprocessor, but in the alternative, the processor may be any commercially available processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, for example, a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration.

[0143] The steps of the method or algorithm described in conjunction with the present disclosure may be implemented directly in hardware, in a software module executed by a processor, or in a combination of the two. The software module may reside in any form of storage medium known in the art. Some examples of usable storage media include random access memory (RAM), read-only memory (ROM), flash memory, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, removable disks, CD-ROMs, and the like. The software module may include a single instruction, or many instructions, and may be distributed over several different code segments, distributed between different programs, and distributed across multiple storage media. A storage medium may be coupled to a processor so that the processor can read and write information from / to the storage medium. In an alternative, a storage medium may be integrated into a processor.

[0144] The disclosed methods include one or more steps or actions for achieving the described methods. These method steps and / or actions may be interchangeable with each other without departing from the scope of the claims. In other words, unless a particular order of steps or actions is specified, the order and / or use of specific steps and / or actions may be changed without departing from the scope of the claims.

[0145] The functions described may be implemented in hardware, software, firmware, or any combination thereof. If implemented in hardware, an example hardware configuration may include a processing system in a device. The processing system may be implemented with a bus architecture. Depending on the specific application and overall design constraints of the processing system, the bus may include any number of interconnecting buses and bridges. The bus may link together various circuits including a processor, a machine-readable medium, and a bus interface. The bus interface may be used to connect a network adapter, etc., to the processing system via the bus, in particular. The network adapter may be used to implement signal processing functions. For some aspects, a user interface (e.g., a keypad, a display, a mouse, a joystick, etc.) may also be connected to the bus. The bus may also link various other circuits, such as timing sources, peripherals, voltage regulators, power management circuits, and similar circuits, which are well known in the art and will not be described further.

[0146] The processor may be responsible for managing the bus and general processing, including executing software stored on a machine-readable medium. The processor may be implemented with one or more general and / or special processors. Examples include microprocessors, microcontrollers, DSP processors, and other circuit systems that can execute software. Software should be broadly interpreted as meaning instructions, data, or any combination thereof, whether referred to as software, firmware, middleware, microcode, hardware description language, or other. As an example, a machine-readable medium may include a random access memory (RAM), a flash memory, a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a register, a disk, an optical disk, a hard drive, or any other suitable storage medium, or any combination thereof. The machine-readable medium may be implemented in a computer program product. The computer program product may include packaging material.

[0147] In hardware implementation, the machine-readable medium can be a part separated from the processor in the processing system. However, as those skilled in the art will readily appreciate, the machine-readable medium or any part thereof can be outside the processing system. As an example, the machine-readable medium may include a transmission line, a carrier wave modulated by data, and / or a computer product separated from the device, all of which can be accessed by the processor through a bus interface. Alternatively or additionally, the machine-readable medium or any part thereof can be integrated into the processor, such as a cache and / or a general register file may be such a situation. Although the various components discussed can be described as having specific locations, such as local components, they can also be configured in various ways, such as some components are configured as a part of a distributed computing system.

[0148] The processing system can be configured as a general processing system having one or more microprocessors providing processor functionality and external memory providing at least a portion of the machine-readable medium, all linked together with other supporting circuitry via an external bus architecture. Alternatively, the processing system can include one or more neuromorphic processors for implementing the described neuron model and neural system model. As another alternative, the processing system can be implemented with an application specific integrated circuit (ASIC) with a processor, bus interface, user interface, supporting circuitry, and at least a portion of the machine-readable medium integrated in a single chip, or with one or more field programmable gate arrays (FPGAs), programmable logic devices (PLDs), controllers, state machines, gated logic, discrete hardware components, or any other suitable circuitry, or any combination of circuits capable of performing the various functionalities described throughout the present disclosure. Depending on the specific application and the overall design constraints imposed on the overall system, those skilled in the art will recognize how to best implement the functionality described with respect to the processing system.

[0149] The machine-readable medium may include several software modules. These software modules include instructions that cause the processing system to perform various functions when executed by the processor. These software modules may include a transmission module and a receiving module. Each software module may reside in a single storage device or be distributed across multiple storage devices. As an example, when a triggering event occurs, the software module can be loaded into the RAM from the hard drive. During the execution of the software module, the processor can load some instructions into the cache to increase the access speed. One or more cache lines can then be loaded into the general register file for execution by the processor. When the functionality of the software module is described below, it will be understood that such functionality is implemented by the processor when the processor executes the instructions from the software module. In addition, it should be appreciated that various aspects of the present disclosure produce improvements in the functionality of processors, computers, machines, or other systems that implement such aspects.

[0150] If implemented in software, each function can be stored on or transmitted by one or more instructions or codes on a computer-readable medium. Computer-readable media include both computer storage media and communication media, and these media include any media that facilitate the transfer of a computer program from one place to another. Storage media can be any available media that can be accessed by a computer. As an example and not limitation, such computer-readable media may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, disk storage or other magnetic storage devices, or any other media that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer. In addition, any connection is also properly referred to as a computer-readable medium. For example, if the software is transmitted from a website, a server, or other remote sources using a coaxial cable, a fiber optic cable, a twisted pair, a digital subscriber line (DSL), or a wireless technology (such as infrared (IR), radio, and microwaves), the coaxial cable, the fiber optic cable, the twisted pair, the DSL, or the wireless technology (such as infrared, radio, and microwaves) are included in the definition of the medium. Disk and disc as used herein include compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk, and Blu-ray Disks, where disks often reproduce data magnetically, and discs reproduce data optically with lasers. Thus, in some aspects, computer-readable media may include non-transitory computer-readable media (e.g., tangible media). Additionally, for other aspects, computer-readable media may include transient computer-readable media (e.g., signals). Combinations of the above should also be included within the scope of computer-readable media.

[0151] Thus, some aspects may include a computer program product for performing the operations presented. For example, such a computer program product may include a computer-readable medium having stored (and / or encoded) thereon instructions that can be executed by one or more processors to perform the described operations. For some aspects, the computer program product may include packaging materials.

[0152] In addition, it should be appreciated that modules and / or other appropriate means for performing the described methods and techniques can be downloaded and / or otherwise obtained by a user terminal and / or base station where applicable. For example, such a device can be coupled to a server to facilitate the transfer of means for performing the described methods. Alternatively, the various methods described can be provided via a storage device (e.g., RAM, ROM, a physical storage medium such as a compact disc (CD) or a floppy disk, etc.) so that once the storage device is coupled to or provided to a user terminal and / or base station, the device can obtain the various methods. In addition, any other suitable technology suitable for providing the described methods and techniques to a device can be utilized.

[0153] It will be understood that the claims are not limited to the precise configuration and components illustrated above. Various changes, substitutions and variations may be made in the arrangement, operation and details of the methods and apparatus described above without departing from the scope of the claims.

Claims

1. A method for classifying an input, comprising: identifying human-object interactions in the input; identifying contextual features of the input; comparing each identified context feature to the identified human-object interaction; determining a value representing a concatenation of the identified contextual features and the identified human-object interaction; as well as The contextual features are fused with the identified human-object interactions to obtain a fused representation when the value is greater than a threshold and the contextual features are not fused with the identified human-object interactions to obtain a representation of the human-object interaction when the value is less than the threshold.

2. The method of claim 1, wherein the contextual features include one or more of clothing, human body part features, human-object spatial relationships, scene location, scene geometry, or co-occurring objects.

3. The method of claim 1, further comprising: The probability of a resulting verb and noun pair is determined based on the representation of the human-object interaction without the contextual feature when the value is less than the threshold, and based on a fused representation of the human-object interaction and the contextual feature when the value is greater than the threshold.

4. The method of claim 1, further comprising adding noise during a training forward pass to enhance value determination.

5. The method of claim 4, further comprising removing the noise during a training back pass.

6. The method of claim 1, wherein the classifier used to classify the input is robust to horizontal and vertical flipping.

7. The method of claim 1, further comprising jittering the position of one or more of the human bounding box or the object bounding box during training.

8. An apparatus comprising: Memory; as well as at least one processor coupled to the memory, the at least one processor configured to: Identify human-object interactions in input; identifying contextual features of the input; comparing each identified context feature to the identified human-object interaction; determining a value representing a concatenation of the identified contextual features and the identified human-object interaction; as well as The contextual features are fused with the identified human-object interactions to obtain a fused representation when the value is greater than a threshold and the contextual features are not fused with the identified human-object interactions to obtain a representation of the human-object interaction when the value is less than the threshold.

9. The apparatus of claim 8, wherein the contextual features include one or more of clothing, human body part features, human-object spatial relationships, scene location, scene geometry, or co-occurring objects.

10. The apparatus of claim 8, wherein the at least one processor is further configured to determine the probability of the resulting verb and noun pair based on the representation of the human-object interaction without the contextual feature when the value is less than the threshold, and based on a fused representation of the human-object interaction and the contextual feature when the value is greater than the threshold.

11. The apparatus of claim 8, wherein the at least one processor is further configured to add noise during a training forward pass to enhance value determination.

12. The apparatus of claim 11, wherein the at least one processor is further configured to remove the noise during a training back pass.

13. The apparatus of claim 8, wherein the classifier used to classify the input is robust to horizontal and vertical flipping.

14. The apparatus of claim 8, wherein the at least one processor is further configured to jitter a position of one or more of a human bounding box or an object bounding box during training.

15. A device comprising: means for identifying human-object interactions in input; means for identifying contextual features of said input; means for comparing each identified contextual feature to the identified human-object interaction; means for determining a value representing a concatenation of the identified contextual feature and the identified human-object interaction; as well as Means for fusing the contextual features with the identified human-object interactions to obtain a fused representation when the value is greater than a threshold and not fusing the contextual features with the identified human-object interactions to obtain a representation of the human-object interaction when the value is less than the threshold.

16. The apparatus of claim 15, wherein the contextual features include one or more of clothing, human body part features, human-object spatial relationships, scene location, scene geometry, or co-occurring objects.

17. The apparatus of claim 15, further comprising: Means for determining a probability of a resulting verb and noun pair based on the representation of the human-object interaction without the contextual feature when the value is less than the threshold, and based on a fused representation of the human-object interaction and the contextual feature when the value is greater than the threshold.

18. The apparatus of claim 15, further comprising means for adding noise to enhance value determination during a training forward pass.

19. The apparatus of claim 18, further comprising means for removing the noise during a training back pass.

20. The apparatus of claim 15, wherein the classifier used to classify the input is robust to horizontal and vertical flipping.

21. The apparatus of claim 15, further comprising means for jittering the position of one or more of a human bounding box or an object bounding box during training.

22. A non-transitory computer readable medium having program code recorded thereon, the program code being executed by a processor and comprising: program code for identifying human-object interactions in input; program code for identifying contextual features of the input; program code for comparing each identified contextual feature to the identified human-object interaction; program code for determining a value representing a concatenation of the identified contextual feature and the identified human-object interaction; as well as Program code for fusing the contextual features with the identified human-object interactions to obtain a fused representation when the value is greater than a threshold and not fusing the contextual features with the identified human-object interactions to obtain a representation of the human-object interaction when the value is less than the threshold.

23. The non-transitory computer-readable medium of claim 22, wherein the contextual features include one or more of clothing, human-object spatial relationships, scene location, scene geometry, or co-occurring objects.

24. The non-transitory computer readable medium of claim 22, further comprising: Program code for determining a probability of a resulting verb and noun pair based on the representation of the human-object interaction without the contextual feature if the value is less than the threshold, and based on a fused representation of the human-object interaction and the contextual feature if the value is greater than the threshold.

25. The non-transitory computer readable medium of claim 22, further comprising program code for adding noise to enhance value determination during a training forward pass.

26. The non-transitory computer readable medium of claim 25, further comprising program code for removing the noise during a training back pass.

27. The non-transitory computer readable medium of claim 22, wherein a classifier for classifying the input is robust to horizontal and vertical flipping.

28. The non-transitory computer readable medium of claim 22, further comprising program code for jittering the position of one or more of a human bounding box or an object bounding box during training.

Images