remove GPU dependency for interference?

[luke4u]

Hi Guys,

First of all, thank you so much for sharing this amazing work. I have run the demo colab and got a good result.

To confirm, to run interference, cuda-enabled GPU is a must?

As #34 mentioned, do you consider to ease the dependency on GPU? This could make the model more scalable.

Thanks again.
Luke

[iiLaurens]

+1 for this. Ensuring a GPU is available in a production environment in the cloud can be a real nuisance. Also since MMdetection 2.0 there is support for CPU-only mode. So if someone is able to reproduce or convert the model to mmdetection 2.0-compatible format, then this model can be used for inference in a CPU-only environment. The nice part is that training can still be done with GPU, but the resulting checkpoints will be able to load and run in a CPU-only environment too.

See also this page on cpu-only mode and this page on upgrading from 1.x to 2.0. Unfortunately I wasn't able to succesfully convert the model myself using the provided conversion tool. Hopefully the creator could help out and provide trained models compatible with mmdetection 2.0.

[iiLaurens]

Since the creator of issue #77 mentioned he was able to convert the model (but unfortunately did not share his config or conversion steps), I decided to give it another shot myself. Succesfully this time.

I would like to refer you all to my branch at iiLaurens/CascadeTabNet:mmdet2x. It includes a demo notebook on how to run using mmdetection v2.3.0 in a cpu only colab environment. You can find that notebook here. All checkpoint files can be found under the releases on this page. Happy inferencing!

[luke4u]

Hi @iiLaurens , thank you for sharing the workflow!
Noticed you are using mmcv-full==1.0.5 There seems no distribution available for the Windows platform in below link, and mmcv-full relies on CUDA? (correct me if I am wrong).

https://openmmlab.oss-accelerate.aliyuncs.com/mmcv/dist/index.html

I had to install mmcv=1.0.5, but ran into an error ModuleNotFoundError: No module named 'mmcv._ext' Btw, do you manage to run the model on a Windows platform with only CPU?

[iiLaurens]

As far as I know there is no windows version for mmcv-full. And as you noticed mmcv simply doesn't work at all. I run in Linux environment.

…

On Fri, Sep 25, 2020, 12:33 Luke ***@***.***> wrote: Hi @iiLaurens <https://github.com/iiLaurens> , thank you for sharing the workflow! Noticed you are using mmcv-full==1.0.5 There seems no distribution available for the Windows platform in below link, and mmcv-full relies on CUDA? (correct me if I am wrong). https://openmmlab.oss-accelerate.aliyuncs.com/mmcv/dist/index.html I had to install mmcv=1.0.5, but ran into an error ModuleNotFoundError: No module named 'mmcv._ext' Btw, do you manage to run the model on a Windows platform with only CPU? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#71 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACLUZ5I2CILVSA5G6BEZIBTSHRWUHANCNFSM4QFKPUUA> .

[kbrajwani]

hi @iiLaurens you only converted the models or after converting you have trained for some epochs?
i am able to convert the model but its output is not perfect as your model.

[iiLaurens]

I did not do any further training, just converting. If my memory serves me correctly, I had to convert both the model and the config file. Did you convert both?

…

On Mon, Nov 2, 2020, 08:32 Kumar Rajwani ***@***.***> wrote: hi @iiLaurens <https://github.com/iiLaurens> you only converted the models or after converting you have trained for some epochs? i am able to convert the model but its output is not perfect as your model. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#71 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACLUZ5OQFT7VPVVCITJHUJLSNZN65ANCNFSM4QFKPUUA> .

[kbrajwani]

No, i have only converted model and i am using config file of mmdetection version 2 which is compatible with the model.

[19debanjanbanerjee98]

How to convert the model and the config file to mmdetection version 2 from version 1?

[kbrajwani]

i have done something like this

import torch
checkpoint = torch.load("/content/epoch_36.pth")

## remove the path which giving error while conversion
checkpoint['meta']['config'] = checkpoint['meta']['config'].replace("/content/drive/My Drive/chunk cascade_mask_rcnn_hrnetv2p_w32_20e.py\n","")

torch.save(checkpoint, "/content/epoch_35.pth")

##convert
!python mmdetection/tools/upgrade_model_version.py /content/epoch_35.pth /content/epoch_37.pth --num-classes 81

##detection
from mmdet.apis import init_detector, inference_detector, show_result_pyplot
import mmcv
# Load model
config_file = '/content/mmdetection/configs/hrnet/cascade_mask_rcnn_hrnetv2p_w32_20e_coco.py'
checkpoint_file = '/content/epoch_37.pth'
model = init_detector(config_file, checkpoint_file, device='cuda:0')

# Test a single image 
img = "/content/5.29.2020 COI - Corvias Construction Partners, LLC_0001.jpg"

# Run Inference
result = inference_detector(model, img)

# Visualization results
show_result_pyplot(model, img, result, score_thr=0.85)

[ashish-kubade]

Since the creator of issue #77 mentioned he was able to convert the model (but unfortunately did not share his config or conversion steps), I decided to give it another shot myself. Succesfully this time.

I would like to refer you all to my branch at iiLaurens/CascadeTabNet:mmdet2x. It includes a demo notebook on how to run using mmdetection v2.3.0 in a cpu only colab environment. You can find that notebook here. All checkpoint files can be found under the releases on this page. Happy inferencing!

@iiLaurens thanks for this effort.
Does that also mean I can use CasCadetabnet architecture with my already installed mmdetection v2.3 even when the network was trained on mmdetection v1.2?

[Trinadhbabu]

@iiLaurens Thank you soo much for your work. only thing I changed to work on my cpu is run this
!pip install mmcv-full==1.0.5 -f https://download.openmmlab.com/mmcv/dist/cpu/torch1.5.0/index.html

instead of this
!pip install mmcv-full==1.0.5+torch1.5.0+cpu -f https://openmmlab.oss-accelerate.aliyuncs.com/mmcv/dist/index.html

[hurshprasad]

I fine-tune/trained a model and was able to upgrade using mmdetection/tools/upgrade_model_version.py , and use @iiLaurens config and run both init_detector, and inference_detector using following package setup.

mmcv-full==1.0.5
mmdet==2.3.0
numpy==1.21.3
opencv-python==4.5.4.58
pycocotools==2.0.2
torch==1.5.1+cpu
torchvision==0.6.1+cpu

However in my CPU version from my checkpoint for inference I get back all empty arrays for the 81 classes. Only difference is that I started General Model table detection link check point and I trained with original config.

If anyone has some ideas of what to try or change would greatly appreciate it.

UPDATE: In case it helps anyone who is also fine-tuning their model, I can't upgrade a model I fine tuned in mmdet 1.2 upgrade it and train w/ mmdet > 2 or infer on CPU from it. I was able to upgrade their checkpoint and train and infer on CPU (I used General Model table detection epoch_24.pth) If it is possible please let me know.

[anhhaibkhn]

@iiLaurens ,
Thank you for your work.
Is this possible to run your notebook or reproduce your result on a local windows environment?
I tried and failed to install the requirements, and it was similar to @luke4u.
If it is not possible to use reproduce on windows, could you share the Linux environment details, or suggest the necessary packages to build a docker file for it.

Thank you for your time, also thanks in advance if anyone could help out with some ideas.

[iiLaurens]

I was able to get it to run from docker container (for use in AWS Lambda). This is the dockerfile:

FROM public.ecr.aws/lambda/python:3.8

RUN yum -y install gcc mesa-libGL

RUN pip install \
  torch==1.6.0+cpu \
  torchvision==0.7.0+cpu \
  -f https://download.pytorch.org/whl/torch_stable.html \
  && rm -rf /root/.cache/pip

RUN pip install \
  mmdet==2.3.0 \
  pycocotools==2.0.2 \
  requests

RUN pip install mmcv-full==1.0.5 -f https://download.openmmlab.com/mmcv/dist/cpu/torch1.6.0/index.html

And you need the converted checkpoint and config files that you can find in my repo.

Then some code like this should make it work:

from mmdet.apis import inference_detector, init_detector

config = '/pdfextract/cascadeTabNet/cascade_mask_rcnn_hrnetv2p_w32_20e.py'
checkpoint = '/pdfextract/cascadeTabNet/General.Model.table.detection.v2.pth'

model = init_detector(config, checkpoint, device='cpu')
results = inference_detector(model, img)

[anhhaibkhn]

Thanks so much for your suggestions.
I will give it a try to build a similar ubuntu container for running it on the local windows.

[anhhaibkhn]

@iiLaurens ,

Thanks a lot again. I just want to let you know that I was able to build a running ubuntu container on windows thanks to your suggestion.

Now, I could get the inference results without any problems on windows with just the CPU. Awesome work!

[mohit-217]

@iiLaurens ,

Thanks a lot again. I just want to let you know that I was able to build a running ubuntu container on windows thanks to your suggestion.

Now, I could get the inference results without any problems on windows with just the CPU. Awesome work!

can you please elaborate your steps

[AGRocky]

Hi Folks I admire the work of @iiLaurens and appreciate the team. However I'm find error as below. Please I request any of you to resolve this issues it would be highly be appreciated. As I'm using colab notebook with cpu
ERROR: Could not find a version that satisfies the requirement torch==1.5.1+cpu (from versions: 1.11.0, 1.11.0+cpu, 1.11.0+cu102, 1.11.0+cu113, 1.11.0+cu115, 1.11.0+rocm4.3.1, 1.11.0+rocm4.5.2, 1.12.0, 1.12.0+cpu, 1.12.0+cu102, 1.12.0+cu113, 1.12.0+cu116, 1.12.0+rocm5.0, 1.12.0+rocm5.1.1, 1.12.1, 1.12.1+cpu, 1.12.1+cu102, 1.12.1+cu113, 1.12.1+cu116, 1.12.1+rocm5.0, 1.12.1+rocm5.1.1, 1.13.0, 1.13.0+cpu, 1.13.0+cu116, 1.13.0+cu117, 1.13.0+cu117.with.pypi.cudnn, 1.13.0+rocm5.1.1, 1.13.0+rocm5.2, 1.13.1, 1.13.1+cpu, 1.13.1+cu116, 1.13.1+cu117, 1.13.1+cu117.with.pypi.cudnn, 1.13.1+rocm5.1.1, 1.13.1+rocm5.2, 2.0.0, 2.0.0+cpu, 2.0.0+cpu.cxx11.abi, 2.0.0+cu117, 2.0.0+cu117.with.pypi.cudnn, 2.0.0+cu118, 2.0.0+rocm5.3, 2.0.0+rocm5.4.2, 2.0.1, 2.0.1+cpu, 2.0.1+cpu.cxx11.abi, 2.0.1+cu117, 2.0.1+cu117.with.pypi.cudnn, 2.0.1+cu118, 2.0.1+rocm5.3, 2.0.1+rocm5.4.2)
ERROR: No matching distribution found for torch==1.5.1+cpu

[mohit-217]

Hey Abhishek this is related to more your dependencies

…

On Tue, 4 Jul, 2023, 11:47 Abhishek G, ***@***.***> wrote: Hi Folks I admire the work of @iiLaurens <https://github.com/iiLaurens> and appreciate the team. However I'm find error as below. Please I request any of you to resolve this issues it would be highly be appreciated. As I'm using colab notebook with cpu ERROR: Could not find a version that satisfies the requirement torch==1.5.1+cpu (from versions: 1.11.0, 1.11.0+cpu, 1.11.0+cu102, 1.11.0+cu113, 1.11.0+cu115, 1.11.0+rocm4.3.1, 1.11.0+rocm4.5.2, 1.12.0, 1.12.0+cpu, 1.12.0+cu102, 1.12.0+cu113, 1.12.0+cu116, 1.12.0+rocm5.0, 1.12.0+rocm5.1.1, 1.12.1, 1.12.1+cpu, 1.12.1+cu102, 1.12.1+cu113, 1.12.1+cu116, 1.12.1+rocm5.0, 1.12.1+rocm5.1.1, 1.13.0, 1.13.0+cpu, 1.13.0+cu116, 1.13.0+cu117, 1.13.0+cu117.with.pypi.cudnn, 1.13.0+rocm5.1.1, 1.13.0+rocm5.2, 1.13.1, 1.13.1+cpu, 1.13.1+cu116, 1.13.1+cu117, 1.13.1+cu117.with.pypi.cudnn, 1.13.1+rocm5.1.1, 1.13.1+rocm5.2, 2.0.0, 2.0.0+cpu, 2.0.0+cpu.cxx11.abi, 2.0.0+cu117, 2.0.0+cu117.with.pypi.cudnn, 2.0.0+cu118, 2.0.0+rocm5.3, 2.0.0+rocm5.4.2, 2.0.1, 2.0.1+cpu, 2.0.1+cpu.cxx11.abi, 2.0.1+cu117, 2.0.1+cu117.with.pypi.cudnn, 2.0.1+cu118, 2.0.1+rocm5.3, 2.0.1+rocm5.4.2) ERROR: No matching distribution found for torch==1.5.1+cpu — Reply to this email directly, view it on GitHub <#71 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AMJEFL6UHR7RW5ASAA4TZ6LXOOYPDANCNFSM4QFKPUUA> . You are receiving this because you commented.Message ID: ***@***.***>

[AGRocky]

Could you please elaborate my friend

[linkstatic12]

Please elaborate more