search

ethnhe / FFB6D

[CVPR2021 Oral] FFB6D: A Full Flow Bidirectional Fusion Network for 6D Pose Estimation.
MIT License
299 stars 71 forks source link

LineMod Multi-object pose estimation? #88

Open nachi9211 opened 2 years ago

nachi9211 commented 2 years ago

FOR LINEMOD DATASET: Is the code able to provide poses for objects irrespective of class label? i.e. can the model be trained without the 'cls' arg.

Or is there existing support in the code for dealing with this?

I can change n_objects in common.py, make adjustments for labels in linemod_dataset.py and use MeanShift.fit_multi_clus, but I am unable to correct it so it trains.

As the code is, is a model only trainable for 1 object class at a time? I presumed I do not need to add a module to handle number of found objects, but now I am not so sure.

nachi9211 commented 2 years ago

Tried using fuse data only with multi-label but did not get good results.