10results about How to "Improve model quality" patented technology

The invention provides a limit test method, system and device for an artificial intelligence model and a medium, and the method specifically comprises the steps: constructing a plurality of scene templates for different types of games, and generating a test case set by combining and transforming the scene templates according to a boundary condition input by a user; based on the test case set, according to a preset optimization objective function, adopting a heuristic search algorithm to determine an optimal test parameter combination meeting the artificial intelligence model; performing automatic testing on the artificial intelligence model according to the optimal test parameter combination, and generating a model test result; and analyzing the model test result, determining the performance of the artificial intelligence model under the boundary condition, generating a model test report, and carrying out optimization and re-test on the artificial intelligence model through the model test report. According to the method, full-process automation and intellectualization of game AI model testing are achieved, and the testing efficiency and the model quality are remarkably improved.

The utility model relates to the field of 3D printing, in particular to a movement device and a 3D printer, and the movement device comprises a first transmission belt extending in the first direction; a second transmission belt extending in the first direction; the third transmission belt extends in the second direction perpendicular to the first direction, the first end of the third transmission belt is in transmission connection with the first transmission belt, and the second end of the third transmission belt is in transmission connection with the second transmission belt; the output assembly is fixed on the third transmission belt; the third transmission belt is configured to drive the output assembly to move in the second direction under the condition that the first transmission belt and the second transmission belt rotate in the same direction; under the condition that the first transmission belt and the second transmission belt rotate reversely, the third transmission belt drives the output assembly to move in the first direction. The problem that the transmission belt is loosened after being used for a long time can be solved, the movement precision of the output assembly is improved, and therefore the printing precision and the model quality are improved.

The utility model relates to a transmission device and a 3D printer, and the transmission device comprises a first transmission belt which extends in the first direction and is provided with a first end and a second end which are located in the first direction, and the relative positions of the first end and the second end of the first transmission belt are fixed; the first tensioning assembly comprises a fixing seat and a pressing block connected to the fixing seat, the fixing seat is fixed to the first transmission belt, and the pressing block is configured to abut against the first transmission belt in the direction forming an included angle with the first direction so as to apply tensioning force to the first transmission belt. The first transmission belt can be effectively tensioned, and the printing precision and the model quality are improved.

The invention relates to the technical field of airborne system design, and discloses an airborne system model modeling rule making method and device, equipment and a medium, which can deeply fuse airborne system design multi-source heterogeneous knowledge, automatically generate executable modeling rules, improve airborne system design efficiency and model quality, and improve airborne system design efficiency and model quality. The problem that modeling rules are inconsistent due to the fact that traditional modeling rules depend on artificial experience is solved, and standardization, consistency and quality control efficiency of airborne system design are improved.

The invention relates to the technical field of three-dimensional model optimization, and discloses a three-dimensional model optimization method based on an indoor transformer substation monitoring video, which comprises the following steps of: establishing a transformer substation model library in which a model is not subjected to mapping processing; arranging a three-dimensional model; acquiring an equipment model in the three-dimensional model, and acquiring a live-action picture of equipment of the same model through a monitoring video; performing mapping processing on a model part of the three-dimensional model by adopting a live-action picture; and completing the construction. Compared with the prior art, the method has the advantages that the modeling speed is higher, local details of all inspection point positions can be covered basically, the inspection point positions can be consistent with real scenes basically, the modeling quality is high, and the modeling efficiency is high. Moreover, the model is constructed by adopting the original monitoring equipment, and a panoramic camera and a laser radar are not needed, so that the construction cost is low.

The application provides a kind of adaptive 3D grid reconstruction and error compensation method for die frame square iron reverse engineering, including using laser scanner to carry out 360 ° scanning on die frame square iron to obtain original point cloud data, and the original point cloud data is registered to the same coordinate system, the original point cloud data is processed by multi-scale bilateral filtering, while constructing point cloud quality evaluation field, identifying data missing area and edge aliasing area, generating point cloud confidence weight map, establishing octree space index and calculating structure tensor to obtain principal curvature value, based on principal curvature and confidence weight, the square iron surface is divided into large plane area, transition feature area and boundary feature area, and the geometric topology skeleton line is extracted.The application decomposes the deviation into systematic error and random error and carries out reverse correction, which not only eliminates the overall deformation error of the workpiece, but also retains the real machining texture, realizes high-quality reverse reconstruction.

This invention provides a fully automated laser scanning detection system and method for tunnels and chambers, belonging to the field of underground space exploration technology. The system includes: a drone, a multi-sensor fusion positioning module, a 3D laser scanner, an adaptive environmental perception and obstacle avoidance module, and a ground control station. The multi-sensor fusion positioning module provides high-precision pose information in environments without GPS signals; the 3D laser scanner collects laser point cloud data of the tunnels and chambers; and the adaptive environmental perception and obstacle avoidance module senses changes in dust and illumination in real time and plans a safe flight path. This invention effectively overcomes the challenges of complex environments such as GPS rejection, uneven illumination, and dust interference within tunnels through multi-source data fusion and adaptive algorithms. It achieves fully automated, high-precision, and high-efficiency 3D topographic detection in areas inaccessible to personnel, providing a reliable technical means for monitoring surrounding rock deformation and assessing risks in goaf areas.

The embodiment of the invention discloses an external ear model edge smoothing method and device based on pipeline cutting and a medium, and relates to the technical field of three-dimensional modeling. The method comprises the steps of obtaining an original external ear model; resampling an edge to be smoothed in the original external ear model according to the average vertex distance of the original external ear model; by taking a new edge formed by connecting the resampling vertexes as a center line, constructing a closed pipeline which extends along the new edge and has the cross section being an equal-radius circular ring; cutting the original external ear model by utilizing the closed pipeline, and reserving a part outside the closed pipeline; and filling the cut surface of the reserved part into a smooth curved surface to obtain a final external ear model. According to the embodiment, the edge smoothing effect of the external ear model can be improved.

The invention relates to a speaking face video generation method based on neural radiation field spatial position optimization. The method comprises the following steps: extracting voice information and a face image area from a reference video; sampling the face image area of the first video frame, and tracking the position of a sampling point on a subsequent video frame; determining N sampling points as two-dimensional key points; defining N three-dimensional space key points in the three-dimensional space, and mapping the three-dimensional space key points to an image plane where the sampling points are located by combining the camera external parameters, the camera internal parameters and the rotation translation parameters of the space points to generate N two-dimensional plane points; calculating Euclidean distances between the N two-dimensional key points and the N two-dimensional plane points, and adjusting related parameters and three-dimensional space key point positions; obtaining a three-dimensional face reconstruction model of the target face; and generating a speaking face video of the target face under the target view angle and the target voice information based on the target view angle, the target voice information and the three-dimensional face reconstruction model of the target face.

The application discloses a question recommendation method based on a schema in a question and answer community, and the steps include: 1, collecting data and constructing a question and answer community network; 2, using a schema mining algorithm to mine the schema in the user network, and constructing a schema network based on the schema; 3, using a network embedding learning algorithm to learn the node embedding of the network, and learning the preferences of the answerer for the questioner and the question; 4, fusing the preferences of the answerer in two aspects, using a scoring function to predict the matching degree of the answerer for a new question, arranging the answerer in descending order according to the matching score, and recommending the first N users to answer the question, so as to complete the question recommendation task. The application combines the schema mining algorithm and the embedding learning algorithm of the schema network, fully captures the historical interaction information of the users in the schema network by using the graph neural network, and fully fuses the personal feature information, network structure information and text semantic information of the users, so that more accurate recommendation effect is realized.