edz-o / 3dnbf Goto Github PK

I'd like to use your pre-trained model to reconstruct videos and images. Could you please guide me on how to test this? I'm very grateful for your assistance.

Thanks for your excellent work.
When I run the demo part, the result is like this:

When I uncommented these two lines of code, and run this command CUDA_VISIBLE_DEVICES=0 python tools/visualize_predictions.py --config configs/3dnbf/resnet50_pare_w_coke_pw3d_demo.py --output_file output/result_keypoints.json --outdir output/visualization again, the result became like this:

I want to know why the human body model is not displayed on my demo results.
Thanks in advance.

Hello, I would like to ask how to replicate the experimental results (the evaluation results in the picture) from your paper? Thank you very much for your response.

Hi @edz-o et al.,

Thx for ur exellent work. May I ask it seems the occ model is never activated for all given configs, no? If I understood correctly, a Bernoulli is used to model occ? Any clarification would help! Thx in advance & best,

As stated in the paper:

but note that we are using a robust likelihood during inference. (Sec. 3.2)

Hi @edz-o @Angtian @AdamKortylewski et al.,

It would be good if u could clarify where 2-stage is implemented in the current codebase. I have trouble localizing it. Thx!

Hi @edz-o et al.,

I could not find details of banks in either the manuscript & supp. Could you elaborate or direct some materials on this? Thx

Hi @edz-o @Angtian @AdamKortylewski et al.,

Could u kindly provide hints on where the NLL loss is computed during training? The loss computation is a bit complicated, and I got confused. Thx!