MMCV instalation loophole between MMCV 1.7.2 and MMCV 2.0.0rc4

[CrisTTT]

Hi everyone.

In trying to run ViTPose according to the instalation and running instructions, I keep banging my head against the same error again and again no matter the environment i use. The issue is the following:
... line 16, in
assert (mmcv_version >= digit_version(mmcv_minimum_version)
AssertionError: MMCV==1.7.2 is used but incompatible. Please install mmcv>=2.0.0rc4, <=3.0.0.

OR:

.local/lib/python3.10/site-packages/mmpose/init.py", line 10, in
mmcv_version = digit_version(mmcv.version)
AttributeError: module 'mmcv' has no attribute 'version'

Currently I am running a WSL Ubuntu 22.04 in Pycharm, I also tried with the normal pycharm venv. Python 3.8 (I also tried 3.7) cuda 11.8, torch 2.1.0 +cu118 torchvision 0.16.0+cu118

I believe there might be something wrong with the mmpose mmcv/mmcv-full versioning/naming?

[ErikValle2]

Check this: #150

[CrisTTT]

Thank you for the indication, I understand I should run that docker image to have the complete correct versions of everything in order to run ViTPose, correct?

[ErikValle2]

Yes. At least it will let you run demos included in this repository. I tried 2d_human_pose_demo.md and it works! However, you must update paths before running any script.

[CrisTTT]

I keep running into this error both on the WSL2 (ubuntu 22.04) and on the venv.

=> ERROR [ 2/11] RUN apt-get update && apt-get install -y git ninja-build libglib2.0-0 libsm6 libxrender-dev libxext6 libgl1-mesa-glx && apt-get clean 3.8s

[ 2/11] RUN apt-get update && apt-get install -y git ninja-build libglib2.0-0 libsm6 libxrender-dev libxext6 libgl1-mesa-glx && apt-get clean && rm -rf /var/lib/apt/lists/*:
...
0.398 The following signatures couldn't be verified because the public key is not available: NO_PUBKEY A4B469963BF863CC
...
2.823 Reading package lists...
3.704 W: GPG error: https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64 InRelease: The following signatures couldn't be verified because the public key is not available: NO_PUBKEY A4B469963BF863CC
3.704 E: The repository 'https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64 InRelease' is not signed.

---

Dockerfile:11

10 |
11 | >>> RUN apt-get update && apt-get install -y git ninja-build libglib2.0-0 libsm6 libxrender-dev libxext6 libgl1-mesa-glx
12 | >>> && apt-get clean
13 | >>> && rm -rf /var/lib/apt/lists/*
14 |

ERROR: failed to solve: process "/bin/sh -c apt-get update && apt-get install -y git ninja-build libglib2.0-0 libsm6 libxrender-dev libxext6 libgl1-mesa-glx && apt-get clean && rm -rf /var/lib/apt/lists/*" did not complete successfully: exit code: 100

Got any clue how i might be able to fix this?

[ErikValle2]

Did you try this: #27? If works, please let me know and I can update the Dockerfile accordingly.

[CrisTTT]

I applied the fix in #27, the docker installed successfully but when running the test command inside the container I got the following:

root@782227951ba0:/mmpose# python demo/topdown_demo_with_mmdet.py demo/mmdetection_cfg/faster_rcnn_r50_fpn_coco.py https://download.openmmlab.com/mmdetection/v2.0/fa
ster_rcnn/faster_rcnn_r50_fpn_1x_coco/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth configs/body_2d_keypoint/topdown_heatmap/coco/td-hm_hrnet-w48_8xb32-210e_coco
-256x192.py https://download.openmmlab.com/mmpose/top_down/hrnet/hrnet_w48_coco_256x192-b9e0b3ab_20200708.pth --input demo/resources/demo.mp4 --output-root vis_results
Traceback (most recent call last):
File "demo/topdown_demo_with_mmdet.py", line 11, in
import mmengine
ModuleNotFoundError: No module named 'mmengine'
root@782227951ba0:/mmpose#

By pip installing mmengine inside the docker, the mmcv version got broken again.

[ErikValle2]

Yes, now I remember I had to fix it manually. Just give me a few minutes and I will update my PR.

[ErikValle2]

This is working fine from my side

ARG PYTORCH="1.12.1"
ARG CUDA="11.3"
ARG CUDNN="8"

FROM pytorch/pytorch:${PYTORCH}-cuda${CUDA}-cudnn${CUDNN}-devel

ENV TORCH_CUDA_ARCH_LIST="6.0 6.1 7.0+PTX"
ENV TORCH_NVCC_FLAGS="-Xfatbin -compress-all"
ENV CMAKE_PREFIX_PATH="$(dirname $(which conda))/../"

RUN rm /etc/apt/sources.list.d/cuda.list

RUN apt-get update && apt-get install -y git ninja-build libglib2.0-0 libsm6 libxrender-dev libxext6 libgl1-mesa-glx\
    && apt-get clean \
    && rm -rf /var/lib/apt/lists/*

# Install xtcocotools
RUN pip install cython
RUN pip install xtcocotools

# Install MMCV
RUN pip install mmcv==2.0.1 -f https://download.openmmlab.com/mmcv/dist/cu113/torch1.12/index.html

# Install MMDet
RUN pip install mmdet mmengine

# Install MMPose
RUN conda clean --all
RUN git clone https://github.com/open-mmlab/mmpose.git /mmpose
WORKDIR /mmpose
RUN mkdir -p /mmpose/data
ENV FORCE_CUDA="1"
RUN pip install -r requirements/build.txt
RUN pip install --no-cache-dir -e .

To run it, just type:

docker build -t vitpose .
docker run -it --name vitpose_test -v $HOME/volume:/volume --gpus all vitpose:latest
python demo/topdown_demo_with_mmdet.py demo/mmdetection_cfg/faster_rcnn_r50_fpn_coco.py https://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/faster_rcnn_r50_fpn_1x_coco/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth configs/body_2d_keypoint/topdown_heatmap/coco/td-hm_hrnet-w48_8xb32-210e_coco-256x192.py https://download.openmmlab.com/mmpose/top_down/hrnet/hrnet_w48_coco_256x192-b9e0b3ab_20200708.pth --input /volume/YOUR_VIDEO.avi --output-root vis_results --save-predictions

Do not forget to mount a volume with a video to test it!

[CrisTTT]

I will give it a try in a couple of hours and let you know of the result, thank you!

[CrisTTT]

The version above gave an error, unfortunately I forgot to keep the log, changed it to the following:
ARG PYTORCH="1.12.1"
ARG CUDA="11.3"
ARG CUDNN="8"

FROM pytorch/pytorch:${PYTORCH}-cuda${CUDA}-cudnn${CUDNN}-devel

Set environment variables

ENV TORCH_CUDA_ARCH_LIST="6.0 6.1 7.0+PTX"
ENV TORCH_NVCC_FLAGS="-Xfatbin -compress-all"
ENV CMAKE_PREFIX_PATH="$(dirname $(which conda))/../"

RUN rm /etc/apt/sources.list.d/cuda.list

Install necessary packages

RUN apt-get update &&
apt-get install -y
perl
git
ninja-build
libglib2.0-0
libsm6
libxrender-dev
libxext6
libgl1-mesa-glx
libssl1.0.0
--fix-missing &&
apt-get clean &&
rm -rf /var/lib/apt/lists/*

Fix broken dependencies (if any)

RUN apt-get update && apt-get install -f && apt-get clean

Install xtcocotools

RUN pip install cython
RUN pip install xtcocotools

Install MMCV from PyPI (Adjust Version Accordingly)

RUN pip install mmcv==2.0.0

Install MMDet

RUN pip install mmdet mmengine

Clone and set up MMPose

RUN conda clean --all
RUN git clone https://github.com/open-mmlab/mmpose.git /mmpose
WORKDIR /mmpose
RUN mkdir -p /mmpose/data
ENV FORCE_CUDA="1"

Install dependencies for MMPose

RUN pip install -r requirements/build.txt
RUN pip install --no-cache-dir -e .

After creating this docker, I tried running the python demo/topdown ... But I got an error related to nms. When checking the cuda versions nvcc gave 11.3 smi gave 12.x.

I decided to abandon the docker idea and go for installing it "manually" to not only build the demo but also be able to fine tune.