Deep Learning Image Processing Parallel Pipeline
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep learning models for image processing are computationally intensive, leading to slow processing speeds during both training and inference phases, particularly in video object detection, where real-time requirements are not met due to sequential computation across channels of layers.
Innovation Solution
Implementing a cache queue in front of each layer of the deep learning model, where a global scheduler divides images into channel data and local schedulers manage computation tasks in parallel, allowing for concurrent processing and storage of results, forming a parallel pipeline computation mode.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If sequential computation is used across channels of layers in deep learning models, then computation accuracy is maintained, but processing speed deteriorates and real-time requirements are not met
Solution Approach 1:
The patent segments the sequential computation process into parallel channels, where each channel processes image data independently simultaneously. The global scheduler divides the image into multiple channel data and distributes them to different computation tasks, enabling parallel processing while maintaining computational accuracy through structured organization of segmented operations.
Solution Approach 2:
The patent transitions from one-dimensional sequential processing to multi-dimensional parallel processing by introducing channel dimensions. Multiple computation tasks operate on different channels concurrently, effectively adding spatial parallelism to the computation architecture, which significantly improves processing speed without compromising accuracy.
2Productivity
If parallel computation tasks are implemented across channels, then processing throughput is improved, but system complexity increases due to scheduling requirements
Solution Approach 1:
The patent introduces a global scheduler as an intermediary that coordinates between multiple computation tasks and cache queues. This intermediary manages the parallel computation workflow by allocating tasks to available resources and synchronizing data flow between layers, enabling high throughput while containing system complexity through centralized coordination.
Solution Approach 2:
The patent implements preliminary action by pre-organizing computation tasks and cache queues before actual processing begins. The global scheduler prepares the computation pipeline in advance, and cache queues are pre-configured to store intermediate results, allowing parallel tasks to execute efficiently without dynamic coordination overhead during computation.
3Productivity
If cache queues are introduced in front of each layer, then parallel processing capability is enhanced, but memory usage increases
Solution Approach 1:
The patent applies local quality by implementing cache queues selectively at specific locations in the computation pipeline - specifically in front of each layer and at intermediate storage positions. This targeted approach provides parallel processing capability where most needed while minimizing overall memory consumption by avoiding unnecessary caching throughout the entire system.
Data Source
AI summary
An image processing method is provided, which is applied to a deep learning model. A cache queue is provided in front of each layer of the deep learning model; a plurality of computation tasks are preset for each layer of the deep learning model in advance, and are configured for computing weight parameters and corresponding to-be-processed data in a plurality of channels in each corresponding layer in parallel, and storing a computation result into a cache queue behind the corresponding layer thereof; in addition, as long as the cache queue in front of the layer includes the computation result stored in the previous layer, the layer can obtain the to-be-processed data from the computation result, subsequent computation is performed, and a parallel pipeline computation mode is also formed between the layers. By means of the mode, the throughput rate during image processing is remarkably improved, and the image processing parallelism degree and speed and the computation performance of the deep learning model are improved. Further provided are an image processing device and system, which have the same beneficial effects as the above image processing method.