kylevedder / zeroflow Goto Github PK

View Code? Open in Web Editor NEW

14.0 2.0 1.0 206 KB

Official repository for ZeroFlow: Scalable Scene Flow via Distillation

Home Page: http://vedder.io/zeroflow

License: MIT License

Python 95.74% Shell 1.35% Jupyter Notebook 2.64% Dockerfile 0.26%

sceneflow

zeroflow's Issues

Got num_samples=0 error when trying to train the model

Thank you for your work.

Following the README and GETTING_STARTED document, I created the train and val supervision labels, and my dataset folder structure follows this:

|-argoverse2
    |-train
    |-train_sceneflow
    |-test
    |-val
    |-val_sceneflow

I mounted the folder to the container correctly as shown below:

root@Max-laptop-debian:/efs# ls
argoverse2
root@Max-laptop-debian:/efs# cd argoverse2/
root@Max-laptop-debian:/efs/argoverse2# ls
test  train  train_sceneflow  val  val_sceneflow

However, when I try to train the model using

python train_pl.py ./configs/fastflow3d/argo/nsfp_distilatation_speed_scaled_updated.py

I always got the num_samples=0 error below

Traceback (most recent call last):
  File "/project/train_pl.py", line 189, in <module>
    main()
  File "/project/train_pl.py", line 146, in main
    train_dataloader = make_train_dataloader(cfg)
  File "/project/train_pl.py", line 68, in make_train_dataloader
    return torch.utils.data.DataLoader(train_dataset,
  File "/miniconda/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 353, in __init__
    sampler = RandomSampler(dataset, generator=generator)  # type: ignore[arg-type]
  File "/miniconda/lib/python3.10/site-packages/torch/utils/data/sampler.py", line 107, in __init__
    raise ValueError("num_samples should be a positive integer "
ValueError: num_samples should be a positive integer value, but got num_samples=0

The paths inside nsfp_distilatation_speed_scaled_updated.py are:

train_sequence_dir = "/efs/argoverse2/train/"
train_flow_dir = "/efs/argoverse2/train_sceneflow/"

test_sequence_dir = "/efs/argoverse2/val/"
test_flow_dir = "/efs/argoverse2/val_sceneflow/"

I believe the problem is that the train path in the config file is incorrect. However, I am unable to find additional information in the document to solve my question. Would you mind pointing out what's wrong with it?

Argoverse 2 Scene Flow generation script link 404

in the getting start ,I need the link https://github.com/nchodosh/argoverse2-sf to compute ground truth flows ,but it's 404.how can i generate the flows?

Questions on supervised traning part code and performance

Thanks for your work and open source,

When I read the code following, I'm wondering why set a assert here for z index must be 0? since Voxelization is 3D, z should not be 0?

zeroflow/models/heads/fast_flow_decoder.py

Lines 15 to 32 in 58a93f7

    
           def forward_single(self, before_pseudoimage: torch.Tensor, 
        
                              after_pseudoimage: torch.Tensor, 
        
                              point_offsets: torch.Tensor, 
        
                              voxel_coords: torch.Tensor) -> torch.Tensor: 
        
               voxel_coords = voxel_coords.long() 
        
               assert (voxel_coords[:, 0] == 0).all(), "Z index must be 0" 
        
               # Voxel coords are Z, Y, X, and the pseudoimage is Channel, Y, X 
        
               # I have confirmed via visualization that these coordinates are correct. 
        
               after_voxel_vectors = after_pseudoimage[:, voxel_coords[:, 1], 
        
                                                       voxel_coords[:, 2]].T 
        
               before_voxel_vectors = before_pseudoimage[:, voxel_coords[:, 1], 
        
                                                         voxel_coords[:, 2]].T 
        
               concatenated_vectors = torch.cat( 
        
                   [before_voxel_vectors, after_voxel_vectors, point_offsets], dim=1) 
        
               flow = self.decoder(concatenated_vectors) 
        
               return flow

kylevedder / zeroflow Goto Github PK

zeroflow's People

Contributors

Stargazers

Watchers

zeroflow's Issues

Got num_samples=0 error when trying to train the model

Argoverse 2 Scene Flow generation script link 404

Questions on supervised traning part code and performance

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent

	def forward_single(self, before_pseudoimage: torch.Tensor,
	after_pseudoimage: torch.Tensor,
	point_offsets: torch.Tensor,
	voxel_coords: torch.Tensor) -> torch.Tensor:
	voxel_coords = voxel_coords.long()
	assert (voxel_coords[:, 0] == 0).all(), "Z index must be 0"

	# Voxel coords are Z, Y, X, and the pseudoimage is Channel, Y, X
	# I have confirmed via visualization that these coordinates are correct.
	after_voxel_vectors = after_pseudoimage[:, voxel_coords[:, 1],
	voxel_coords[:, 2]].T
	before_voxel_vectors = before_pseudoimage[:, voxel_coords[:, 1],
	voxel_coords[:, 2]].T
	concatenated_vectors = torch.cat(
	[before_voxel_vectors, after_voxel_vectors, point_offsets], dim=1)

	flow = self.decoder(concatenated_vectors)
	return flow