AttributeError and performance of MVTN

[JP4CK3T]

AttributeError and performance of MVTN

I want to test that 3D shape retrieval mAP 89.2 or higher is measured utilizing MVTN.

The following environment was configured and the code published on gituhub was executed. (Using GPU RTX 3090)

$ conda create -y -n MVTN python=3.7
$ conda activate MVTN
$ conda install pytorch=1.11.0 torchvision torchaudio cudatoolkit=11.3 -c pytorch
$ conda install -c fvcore -c iopath -c conda-forge fvcore iopath
$ conda install -c bottler nvidiacub
$ conda install pytorch3d -c pytorch3d
$ conda install pandas
$ conda install -c conda-forge trimesh
$ pip install einops imageio scipy matplotlib tensorboard h5py metric-learn

start of training

$ python3 run_mvtn.py --data_dir data/ModelNet40/ --run_mode train --mvnetwork mvcnn --nb_views 12 --views_config learned_spherical

start of shape retrieval

$ python3 run_mvtn.py --data_dir data/ModelNet40/ --run_mode train --mvnetwork mvcnn --nb_views 12 --views_config learned_spherical

As a result(Epoch = 100) of the execution, shape retrieval mAP was measured as 0.609...

I did two main things to solve the error that occurred while running the code.

1. Fix bin_size in Pytorch3d = 0
2. Remove .item() method due to AttributeError: 'float' object has no attribute 'item'

Excuse me, can you give me some hints to achieve my shape retrieval mAP goal?

Thank you for your research and look forward to your kind reply.