Upcoming Changes
10C Winter praktikum:
- Quantitative testing. YCB objects dataset: use their trained model and also retrain DOPE on the same YCB data you feed to your model. This way you can also quantitatively show how nice their YCB finetuning was to get the best results they could.
Our own dataset: use our tool to generate data for both dope and ours.
-
Multiple objects (might be easy)
-
Multi camera (look at anipose implementation)
-
Tracking (I/you talk with Cheng about this)
Compare ours/ours+tracking with dope
-
Come up with a good name for your synthetic generator tool.
-
Come up with a good name for our deeplabcut/anipose based 6d pose estimator.