edgar-tr / patchnets Goto Github PK

I tried to set [use_precomputed]=false but error still happened cause of ：

if sdf_samples_with_normals is None:
    if use_precomputed_init:
        return torch.from_numpy(np.load(initialization_file))  
    else:
        npz = np.load(sdf_filename)
        pos_tensor = deep_sdf.data.remove_nans(torch.from_numpy(np.concatenate([npz["pos"], npz["pos_normals"]], axis=1))).numpy()
        neg_tensor = deep_sdf.data.remove_nans(torch.from_numpy(np.concatenate([npz["neg"], npz["neg_normals"]], axis=1))).numpy()

#----------

import numpy as np
cat=np.load('/user-data/patchnets-master/code/data/SdfSamples/ShapeNetV2/02691156/1a29042e20ab6f005e9e2656aff7dd5b.npz')
cat.files
['pos', 'neg']

Hi, thanks for your work.

I want to reproduce this code on my machine, but I find no python environment dependency files in this repo, like environment.yml or requirements.txt. So I wonder how to prepare a satisfied environment? I can not find these files in the repo of DeepSDF either.

Another question is that how to configure the dataset? Which dataset I need to download and where should I put them in order to train the network by my own?

like 【teaser.png】https://raw.githubusercontent.com/edgar-tr/patchnets/master/misc/teaser.png
Hi，sorry to trouble you . i wonder is it posiible to get the image from preoprocessed .npz file directly（through point cloud processing software）？If not, How to get the image from latent code,or by other way? Thank you!

I wanted to get grad of input xyz，but function returned None，and requires_grad=True/certain_grad doesnot work.Is it because the input of network concate x and xyz?
def patch_network_forward(self, input,):
xyz = input[:, -original_coordinate_size:]
...,
for layer in range(0, self.num_layers - 1):
lin = getattr(self, "patch_lin" + str(layer))
if layer in self.latent_in:
x = torch.cat([x, input], 1)
elif layer != 0 and self.xyz_in_all:
x = torch.cat([x, xyz], 1)
x = lin(x)
if layer < self.num_layers - 2:
if (
self.norm_layers is not None
and layer in self.norm_layers
and not self.weight_norm
):
bn = getattr(self, "patch_bn" + str(layer))
x = bn(x)
#x = self.softplus(x)
x = self.relu(x)
#x = self.elu(x)
if self.dropout is not None and layer in self.dropout:
x = F.dropout(x, p=self.dropout_prob, training=self.training)

Hello,

Thank you for releasing this amazing work, by the way, could you kindly release your pretrained model so others can evaluate your method rather than training from scratch?

Best.

Hi,
an error happened when I tried to use evaluate_patch_network_metrics() on useful_scripts.py
File "useful_scripts.py", line 1702, in main
evaluate_patch_network_metrics()
File "useful_scripts.py", line 1141, in evaluate_patch_network_metrics
visualize_mixture(results_folder=results_folder, grid_resolution=grid_resolution, mesh_files=evaluate_json, data_source=data_source, break_if_latent_does_not_exist=True, output_name=regressed_meshes_folder,checkpoints=3)
File "useful_scripts.py", line 55, in visualize_mixture
network, patch_latent_size, mixture_latent_size, load_weights_into_network, sdf_to_latent, patch_forward, latent_to_mesh, get_training_latent_vectors, specs = _setup_torch_network(results_folder,checkpoint=checkpoints)
File "useful_scripts.py", line 135, in _setup_torch_network
network = Network(patch_latent_size=patch_latent_size,
File "/root/autodl-tmp/patchnets_neuralpull/code/networks/deep_sdf_decoder.py", line 357, in init
self._init_patch_network_training(train_patch_network, patch_network_pretrained_path, results_folder)
File "/root/autodl-tmp/patchnets_neuralpull/code/networks/deep_sdf_decoder.py", line 404, in _init_patch_network_training
self.load_state_dict(current_weights)
File "/root/miniconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1223, in load_state_dict
raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for Decoder:
size mismatch for patch_lin0.weight_v: copying a param with shape torch.Size([128, 35]) from checkpoint, the shape in current model is torch.Size([128, 131]).
size mismatch for patch_lin3.bias: copying a param with shape torch.Size([224]) from checkpoint, the shape in current model is torch.Size([128]).
size mismatch for patch_lin3.weight_g: copying a param with shape torch.Size([224, 1]) from checkpoint, the shape in current model is torch.Size([128, 1]).
size mismatch for patch_lin3.weight_v: copying a param with shape torch.Size([224, 128]) from checkpoint, the shape in current model is torch.Size([128, 128]).
I am confusued .The network recorded between specs train.json and test.json are identical. How to fix it?

I noticed that some .obj files of ShapeNetV2 cannot be correctly evaluateed. They can be generated when [generate_meshes = True] on useful_scripts.py ,but an error will happen when evaluate_patch_network_metrics runs again with [generate_meshes = False]

Traceback (most recent call last):
File "/root/miniconda3/lib/python3.8/multiprocessing/pool.py", line 125, in worker
result = (True, func(*args, **kwds))
File "/root/miniconda3/lib/python3.8/multiprocessing/pool.py", line 48, in mapstar
return list(map(*args))
File "useful_scripts.py", line 1101, in _evaluate_object
iou = intersection_over_union(groundtruth_mesh_file, regressed_trimesh)
File "useful_scripts.py", line 1028, in intersection_over_union
watertight_result = precompute_iou_on_watertight(groundtruth_mesh_file, num_points=num_points) # num_points x 4 (x,y,z, 1.0f if inside else 0.0f)
File "useful_scripts.py", line 1013, in precompute_iou_on_watertight
max_bb, min_bb = _get_bounding_box(groundtruth_trimesh.vertices)
AttributeError: 'Scene' object has no attribute 'vertices'

Is it because these objects are nor water-tight? And how to fix it?

Hello, thanks to your amazing work!

Should I get pre-trained weights for skipping training and directly conducting evaluation?

initialization_file = sdf_filename + "init" + str(patch_latent_size) + "" + str(num_patches) + "" + str(surface_sdf_threshold) + "_" + str(final_scaling_increase_factor) + ("_tiny" if use_tiny_patchnet else "") + ".npy"

but there's no such file or directory: 'datasets/shapenet_v2/preprocessed/SdfSamples/ShapeNetV2/04554684/1a23fdbb1b6d4c53902c0a1a69e25bd9.npz_init_128_30_0.02_1.2.npy'
(only have 1a23fdbb1b6d4c53902c0a1a69e25bd9.npz)