search

nickgkan / 3d_diffuser_actor

Code for the paper "3D Diffuser Actor: Policy Diffusion with 3D Scene Representations"
https://3d-diffuser-actor.github.io/
MIT License
159 stars 16 forks source link

Dependency Issues #21

Closed MaximilianToelle closed 2 months ago

MaximilianToelle commented 2 months ago

Hey,

Thanks for your amazing work! I want to point out a current installation dependency problem:

pip install dgl -f https://data.dgl.ai/wheels/cu116/dgl-1.1.3%2Bcu116-cp38-cp38-manylinux1_x86_64.whl installs dgl version 2.1.0 and also torch version 2.2.2. This in itself is a problem as dgl version 2.1.0 only supports torch versions 2.0.0/1, 2.1.0/1/2, 2.2.0/1.

On the other hand, your environment.yaml file depends on pytorch 1.13.1, which does not work with dgl 2.1.0. In addition, pytorch 1.13.1 does not work with torchdata 0.7.1. However, it seems that you need torchdata>0.5.1, which is the last version that is compatible with pytorch 1.13.1.

What worked for me was to install torch 2.1.0 and update torchvision to 0.16 after having installed the newest dgl version using your command. Maybe your intent was to install dgl 1.1.3, which supports pytorch 1.13.1, but I think your pip install link should then be https://data.dgl.ai/wheels/dgl-1.1.3-cp38-cp38-manylinux1_x86_64.whl.

pawanw17 commented 2 months ago

I have faced this issue too... Ended up using dgl v2.1.0 with some workarounds here and there

EDIT: Adding a log

(3d_diffuser_actor_new) pawan.wadhwani@phoenix:~/3d_diffuser_actor$ pip install dgl -f https://data.dgl.ai/wheels/cu116/dgl-1.1.3%2Bcu116-cp38-cp38-manylinux1_x86_64.whl
Looking in links: https://data.dgl.ai/wheels/cu116/dgl-1.1.3%2Bcu116-cp38-cp38-manylinux1_x86_64.whl
WARNING: Skipping page https://data.dgl.ai/wheels/cu116/dgl-1.1.3%2Bcu116-cp38-cp38-manylinux1_x86_64.whl because the HEAD request got Content-Type: binary/octet-stream. The only supported Content-Types are application/vnd.pypi.simple.v1+json, application/vnd.pypi.simple.v1+html, and text/html
Collecting dgl
  Using cached dgl-2.1.0-cp38-cp38-manylinux1_x86_64.whl.metadata (581 bytes)

EDIT 2: The right command should be

 pip install dgl==1.1.3+cu116 -f https://data.dgl.ai/wheels/cu116/dgl-1.1.3%2Bcu116-cp38-cp38-manylinux1_x86_64.whl
twke18 commented 2 months ago

Hi,

The EDIT 2 suggested by @pawanw17 works! Thanks for the suggestion and we will update the README accordingly.