ValueError: You need to provide a valid (batch_size)x3x4 projection matrix

[marcomameli1992]

After the solved installation problem with the use of a docker image based on CUDA 10 and some changes on the cpp code of the neural renderer the code start, but when I try to use the entrypoint_predict I receive the next error:

python3.8 entrypoint_predict.py --options ./experiments/default/resnet.yml --checkpoint ./datasets/data/pretrained/resnet50-19c8e357.pth --folder ./datasets/imgs --name test1
=> creating logs/test1
=> creating checkpoints/test1/default_resnet_0505062026
=> creating summary/test1/default_resnet_0505062026
{'checkpoint': './datasets/data/pretrained/resnet50-19c8e357.pth',
 'checkpoint_dir': 'checkpoints/test1/default_resnet_0505062026',
 'dataset': {'camera_c': array([111.5, 111.5]),
             'camera_f': array([248., 248.]),
             'mesh_pos': array([ 0. ,  0. , -0.8]),
             'name': 'shapenet_demo',
             'normalization': True,
             'num_classes': 13,
             'predict': {'folder': './datasets/imgs'},
             'shapenet': {'num_points': 9000,
                          'resize_with_constant_border': False},
             'subset_eval': 'test_tf',
             'subset_train': 'train_tf'},
 'log_dir': 'logs/test1',
 'log_level': 'info',
 'loss': {'weights': {'chamfer': array([1., 1., 1.]),
                      'chamfer_opposite': 0.55,
                      'constant': 1.0,
                      'edge': 0.1,
                      'laplace': 0.5,
                      'move': 0.033,
                      'normal': 0.00016,
                      'reconst': 0.0}},
 'model': {'align_with_tensorflow': False,
           'backbone': 'resnet50',
           'coord_dim': 3,
           'gconv_activation': True,
           'hidden_dim': 192,
           'last_hidden_dim': 192,
           'name': 'pixel2mesh',
           'z_threshold': 0},
 'name': 'test1',
 'num_gpus': 8,
 'num_workers': 16,
 'optim': {'adam_beta1': 0.9,
           'lr': 0.0001,
           'lr_factor': 0.3,
           'lr_step': array([30, 70, 90]),
           'name': 'adam',
           'sgd_momentum': 0.9,
           'wd': 1e-06},
 'pin_memory': True,
 'summary_dir': 'summary/test1/default_resnet_0505062026',
 'test': {'batch_size': 8,
          'dataset': array([], dtype=float64),
          'shuffle': False,
          'summary_steps': 50,
          'weighted_mean': False},
 'train': {'batch_size': 8,
           'checkpoint_steps': 10000,
           'num_epochs': 110,
           'shuffle': True,
           'summary_steps': 50,
           'test_epochs': 1,
           'use_augmentation': True},
 'version': 'default_resnet_0505062026'}
=> creating summary writer
Using GPUs: [0, 1, 2, 3, 4, 5, 6, 7]
Loading datasets: shapenet_demo
Running model initialization...
Traceback (most recent call last):
  File "entrypoint_predict.py", line 39, in <module>
    main()
  File "entrypoint_predict.py", line 34, in main
    predictor = Predictor(options, logger, writer)
  File "/home/vrai/Pixel2Mesh/functions/predictor.py", line 20, in __init__
    super().__init__(options, logger, writer, training=False, shared_model=shared_model)
  File "/home/vrai/Pixel2Mesh/functions/base.py", line 55, in __init__
    self.init_fn(shared_model=shared_model)
  File "/home/vrai/Pixel2Mesh/functions/predictor.py", line 42, in init_fn
    self.renderer = MeshRenderer(self.options.dataset.camera_f, self.options.dataset.camera_c,
  File "/home/vrai/Pixel2Mesh/utils/vis/renderer.py", line 36, in __init__
    self.renderer = nr.Renderer(camera_mode='projection',
  File "/usr/local/lib/python3.8/dist-packages/neural_renderer-1.1.3-py3.8-linux-x86_64.egg/neural_renderer/renderer.py", line 34, in __init__
    raise ValueError('You need to provide a valid (batch_size)x3x4 projection matrix')
ValueError: You need to provide a valid (batch_size)x3x4 projection matrix

I use the default configuration in the default folder and I also tested with the configuration in the baseline folder, after the changes to the baseline from 8 GPUs to 1 (I know that it is not enough but for testing, I use my personal PC)

[ultmaster]

I'll give you some clue:

• You mentioned you are using 1 GPU, why is this in your log: Using GPUs: [0, 1, 2, 3, 4, 5, 6, 7]
• Try to found out the matrix shape in /usr/local/lib/python3.8/dist-packages/neural_renderer-1.1.3-py3.8-linux-x86_64.egg/neural_renderer/renderer.py, and trace where it comes from. See why it's invalid.

[marcomameli1992]

Dear How can I trace the provenience of the matrix?

[WEI18William]

Hi, have you addressed this issue?

Dear How can I trace the provenience of the matrix?