I am working with DLC2.2b8 and have a Key Error when cross validating and trying to convert to tracklets, after retraining the network.
Previously, I had trained/evaluated the network and analyzed a video. The video created from the full pickle file did an awesome job at labeling the desired points! But then I ran into the same problem that I have seen a few people post about on this forum, where the markers disappear when converting to tracklets. I followed suggestions in those posts, and by adjusting some parameters in the inference_cfg.yaml I was able to get some tracklets, but still not as good as the video from the full pickle file. So I decided to add in more labeled frames and retrain the network.
Retraining and evaluating went smoothly. Then I get this error from cross validating:
I wasn’t sure what it was, so moved on to analyzing a video. The analysis was fine and again made a great video with the full pickle file. Next, converting to tracklets gives me this error:
I think this might come from something I have changed in the inference_cfg.yaml file, but I’m not sure what. I tried changing the parameters back close to defaults shown in the GitHub docs and retrying, but this gave me the same errors. My inference file looks like this right now:
Thank you so much for your help!