Video retrieval method and apparatus, device, and storage medium
View Patent ↗This application provides a video retrieval method performed by a computer device. The method includes: performing feature extraction on an image feature of a query video to obtain a first quantization feature, obtaining a second candidate video with a high category similarity to the query video based on the first quantization feature, and finally taking a second candidate video with a high content similarity to the query video as a target video. The quantization control parameters are adjusted according to the texture feature loss value corresponding to each training sample to cause the target quantization processing sub-model to learn the ranking ability of the target texture feature sub-model, to ensure that the ranking effect of two sub-models tend to be consistent, and an end-to-end model architecture enables the target quantization processing sub-model to obtain the corresponding quantization feature based on the image feature.
1. A video retrieval method, performed by a computer device, the method comprising:
receiving a search request from a terminal device, the search request including a query video submitted by a user through a video retrieval interface;
performing feature extraction on the query video using a target image processing sub-model of a trained target video retrieval model to obtain a corresponding image feature of the query video;
performing feature extraction on the image feature using a target quantization processing sub-model of the target video retrieval model to obtain a corresponding first quantization feature;
identifying, from first candidate videos, at least one second candidate video whose associated category similarity to the query video meets a set category similarity requirement based on the first quantization feature;
identifying, among the at least one second candidate video, a target video whose associated content similarity to the query video meets a set content similarity requirement, further including:
determining a ratio between a total matching duration and a comparison duration as a content repetition degree between the query video and one second candidate video;
determining that a content similarity between the query video and the one second candidate video meets the set content similarity requirement when the content repetition degree exceeds a set content repetition degree threshold value; and
identifying the one second candidate video as the target video; and
returning the target video to the terminal device, wherein the target video is configured to be presented on the video retrieval interface and a progress bar of the target video includes at least one segment that is deemed to be similar to the query video and marked visually different from the rest of the target video.
2. The method according to claim 1 , wherein the total matching duration is determined as a matching duration between the at least one second candidate video and the query video, and the comparison duration is determined as a duration value of a shorter video duration between the query video and the one second candidate video.
3. The method according to claim 1 , wherein the identifying, from first candidate videos, at least one second candidate video whose associated category similarity to the query video meets a set category similarity requirement based on the first quantization feature comprises:
determining a quantization feature distance between the first quantization feature and a second quantization feature of each of the first candidate videos; and
determining a first candidate video with a quantization feature distance lower than a preset quantization feature distance threshold value as a second candidate video, each second quantization feature characterizing a video category to which a corresponding at least one first candidate video belongs.
4. The method according to claim 1 , wherein a quantization control parameter of the target quantization processing sub-model is adjusted based on a texture feature loss value corresponding to each training sample during a training process.
5. The method according claim 4 , wherein the texture feature loss value is determined based on a texture control parameter preset for a texture processing sub-model to be trained during a parameter adjustment of the texture processing sub-model to be trained.
6. A computer device comprising a processor and a memory, the memory storing program codes that, when executed by the processor, cause the computer device to perform a video retrieval method including:
receiving a search request from a terminal device, the search request including a query video submitted by a user through a video retrieval interface;
performing feature extraction on the query video using a target image processing sub-model of a trained target video retrieval model to obtain a corresponding image feature of the query video;
performing feature extraction on the image feature using a target quantization processing sub-model of the target video retrieval model to obtain a corresponding first quantization feature;
identifying, from first candidate videos, at least one second candidate video whose associated category similarity to the query video meets a set category similarity requirement based on the first quantization feature;
identifying, among the at least one second candidate video, a target video whose associated content similarity to the query video meets a set content similarity requirement, further including:
determining a ratio between a total matching duration and a comparison duration as a content repetition degree between the query video and one second candidate video;
determining that a content similarity between the query video and the one second candidate video meets the set content similarity requirement when the content repetition degree exceeds a set content repetition degree threshold value; and
identifying the one second candidate video as the target video; and
returning the target video to the terminal device, wherein the target video is configured to be presented on the video retrieval interface and a progress bar of the target video includes at least one segment that is deemed to be similar to the query video and marked visually different from the rest of the target video.
7. The computer device according to claim 6 , wherein the total matching duration is determined as a matching duration between the at least one second candidate video and the query video, and the comparison duration is determined as a duration value of a shorter video duration between the query video and the one second candidate video.
8. The computer device according to claim 6 , wherein the identifying, from first candidate videos, at least one second candidate video whose associated category similarity to the query video meets a set category similarity requirement based on the first quantization feature comprises:
determining a quantization feature distance between the first quantization feature and a second quantization feature of each of the first candidate videos; and
determining a first candidate video with a quantization feature distance lower than a preset quantization feature distance threshold value as a second candidate video, each second quantization feature characterizing a video category to which a corresponding at least one first candidate video belongs.
9. The computer device according to claim 6 , wherein a quantization control parameter of the target quantization processing sub-model is adjusted based on a texture feature loss value corresponding to each training sample during a training process.
10. The computer device according to claim 9 , wherein the texture feature loss value is determined based on a texture control parameter preset for a texture processing sub-model to be trained during a parameter adjustment of the texture processing sub-model to be trained.
11. A non-transitory computer readable storage medium storing program codes that, when executed by a processor of a computer device, cause the computer device to perform a video retrieval method including:
receiving a search request from a terminal device, the search request including a query video submitted by a user through a video retrieval interface;
performing feature extraction on the query video using a target image processing sub-model of a trained target video retrieval model to obtain a corresponding image feature of the query video;
performing feature extraction on the image feature using a target quantization processing sub-model of the target video retrieval model to obtain a corresponding first quantization feature;
identifying, from first candidate videos, at least one second candidate video whose associated category similarity to the query video meets a set category similarity requirement based on the first quantization feature;
identifying, among the at least one second candidate video, a target video whose associated content similarity to the query video meets a set content similarity requirement, further including:
determining a ratio between a total matching duration and a comparison duration as a content repetition degree between the query video and one second candidate video;
determining that a content similarity between the query video and the one second candidate video meets the set content similarity requirement when the content repetition degree exceeds a set content repetition degree threshold value; and
identifying the one second candidate video as the target video; and
returning the target video to the terminal device, wherein the target video is configured to be presented on the video retrieval interface and a progress bar of the target video includes at least one segment that is deemed to be similar to the query video and marked visually different from the rest of the target video.
12. The non-transitory computer readable storage medium according to claim 11 , wherein the total matching duration is determined as a matching duration between the at least one second candidate video and the query video, and the comparison duration is determined as a duration value of a shorter video duration between the query video and the one second candidate video.
13. The non-transitory computer readable storage medium according to claim 11 , wherein a quantization control parameter of the target quantization processing sub-model is adjusted based on a texture feature loss value corresponding to each training sample during a training process.
14. The non-transitory computer readable storage medium according to claim 13 , wherein the texture feature loss value is determined based on a texture control parameter preset for a texture processing sub-model to be trained during a parameter adjustment of the texture processing sub-model to be trained.
15. The non-transitory computer readable storage medium according to claim 11 , wherein the identifying, from first candidate videos, at least one second candidate video whose associated category similarity to the query video meets a set category similarity requirement based on the first quantization feature comprises:
determining a quantization feature distance between the first quantization feature and a second quantization feature of each of the first candidate videos; and
determining a first candidate video with a quantization feature distance lower than a preset quantization feature distance threshold value as a second candidate video, each second quantization feature characterizing a video category to which a corresponding at least one first candidate video belongs.