Using Detectron2 on CPU

[jan-schaeffer]

Hi,

first of all, big thanks for this project!

I am currently working on somethng very similar to the BibItem example you provide here.

Unfortunately, I am having an issue with running the train_net.py. Since I am on a M1 Mac, I do not have CUDA support, though your script assumes it. This causes the following error:

File "/opt/miniconda3/envs/lp/lib/python3.10/site-packages/torch/cuda/__init__.py", line 211, in _lazy_init raise AssertionError("Torch not compiled with CUDA enabled") AssertionError: Torch not compiled with CUDA enabled

Is there any way to change this? I found this and this online, though I am not sure on where to add this in the script.

Any help would be highly appreciated since I need to train a custom model for a work project.

Thanks,

Jan

[cmeer]

I have the same problem

[cmeer]

I got a step further.

In the file train_net.py just above the lines (around line 138): cfg.freeze() default_setup(cfg, args)

you have to put: cfg.MODEL.DEVICE = 'cpu'

After that i had to modify my train_bib.sh file to. python train_net.py
--dataset_name bib-item
--json_annotation_train ../data/bib/train.json
--image_path_train ../data/bib/
--json_annotation_val ../data/bib/test.json
--image_path_val ../data/bib/
--config-file ../configs/prima/fast_rcnn_R_50_FPN_3x.yaml
OUTPUT_DIR ../outputs/bib/fast_rcnn_R_50_FPN_3x/
SOLVER.IMS_PER_BATCH 2

From mask_rcnn to fast_rcnn.

I followed https://www.youtube.com/watch?v=puOKTFXRyr4 this tutorial.

It seems to be training now.

[rodrigo1990]

I got a step further.

In the file train_net.py just above the lines (around line 138): cfg.freeze() default_setup(cfg, args)

you have to put: cfg.MODEL.DEVICE = 'cpu'

After that i had to modify my train_bib.sh file to. python train_net.py --dataset_name bib-item --json_annotation_train ../data/bib/train.json --image_path_train ../data/bib/ --json_annotation_val ../data/bib/test.json --image_path_val ../data/bib/ --config-file ../configs/prima/fast_rcnn_R_50_FPN_3x.yaml OUTPUT_DIR ../outputs/bib/fast_rcnn_R_50_FPN_3x/ SOLVER.IMS_PER_BATCH 2

From mask_rcnn to fast_rcnn.

I followed https://www.youtube.com/watch?v=puOKTFXRyr4 this tutorial.

It seems to be training now.

also it works for me !