Hello, this is team 1885. We created a successful and accurate model for our red prop, but when we try to replicate this model for the blue prop, our training fails relatively early on. We are using the same number of videos and a similar amount of variations of light, angles, and positions. Can anyone assist us?
Before I go through a pretty involved analysis, can you be more specific about what you mean by “fails relatively early on?”
Also, have you read these previous forum threads that deal with similar issues?
Yes, our training runs for around 10-30 mins(Around 1000 steps) and then unexpectantly fails.
It seems that we have solved the issue. We had a few ignored frames that we originally didn’t think were clean/nice enough to use. But this is why our training was failing. We fixed those ignored frames and the model was successfully trained.