Adding Action Chunking with Transformers (ACT) to baselines #640
Conversation
|
before I review code, make sure to add a README.md similar to the other baselines, just use DP as a reference (how to setup the conda/mamba env, citation etc.) |
| from torchvision.models._utils import IntermediateLayerGetter | ||
| from typing import Dict, List | ||
|
|
||
| from ..utils import NestedTensor, is_main_process |
There was a problem hiding this comment.
use absolute imports when possible, it is just the style choice this repo uses.
There was a problem hiding this comment.
I changed them to absolute imports, but I'm not sure if they are correct. please let me know if they need to be fixed. I also added a README file.
|
|
||
| ```bash | ||
| conda create -n act-ms python=3.9 | ||
| conda activate act-ms |
There was a problem hiding this comment.
so the conda env act-ms is created and you do a local pip install. However a simple setup.py file is still missing, can you create that?
| from torchvision.models._utils import IntermediateLayerGetter | ||
| from typing import Dict, List | ||
|
|
||
| from examples.baselines.act.act.utils import NestedTensor, is_main_process |
There was a problem hiding this comment.
imports should be absolute and relative to act (which you pip install -e .)
| import torch | ||
| from torch import nn | ||
|
|
||
| from examples.baselines.act.act.utils import NestedTensor |
|
Final thing probably, can you create an examples.sh script for users to run? Should contain 2 scripts for a few envs, 1 for demonstration loading/replaying and 1 for training. Maybe just 2 for state based and 2 exampls for RGBD that work okay is fine. |
replaced by train_rgbd.py
#617