we are using LEAP to track single points in videos of moving animals and have encountered the issue of large jumps of tracked coordinates (using the LEAP standard setup) in nearby time frames, easily spotted by eye.
The trajectory (see attachment for an example, large jumps occur in the left part of the picture) was obtained from a different training video.
Are there any experiences with such a problem? We thought about re-training the network for every video or simply training longer.
Dear all,
we are using LEAP to track single points in videos of moving animals and have encountered the issue of large jumps of tracked coordinates (using the LEAP standard setup) in nearby time frames, easily spotted by eye.
The trajectory (see attachment for an example, large jumps occur in the left part of the picture) was obtained from a different training video.
Are there any experiences with such a problem? We thought about re-training the network for every video or simply training longer.
Thanks a lot for your help.
Best, Wilhelm