How to reproduce

Question

How to reproduce

lfgogogo opened this issue 2 years ago · 4 comments

Hi,thank you for your brilliant work.May i ask how to reproduce the log here using the open source models ? Looking forward to your reply.

Answer 1 · 2023-06-28T04:03:56.000Z

Go to the README and find and follow the section "Training".
using the command:
CUDA_VISIBLE_DEVICES=0 python train_net.py --config-file configs/{}-segmentation/PADing.yaml --num-gpus 1 MODEL.WEIGHTS pretrained_weight_{}.pth {} can be semantic,instance and panoptic.

I hope it is helpful to you!

Answer 2 · 2023-06-28T07:44:10.000Z

Thanks,i tried,but i got a warning :Some model parameters or buffers are not found in the checkpoint,
then a error:RuntimeError: CUDA error: device-side assert triggered,it seems the model weight does not match with the config file ,and the command was CUDA_VISIBLE_DEVICES=0 python train_net.py --config-file configs/semantic-segmentation/PADing.yaml --num-gpus 1 MODEL.WEIGHTS pretrained_weight_semantic.pth.By the way,i ran the inference command which works fine.

Answer 3 · 2023-06-28T09:35:48.000Z

Sorry to bother,one more question,how can i visualize the results in the json?I just print the segmentation,i get something like this:

What does it mean?

Answer 4 · 2023-06-28T11:05:56.000Z

You can refer to https://github.com/facebookresearch/detectron2/blob/main/tools/visualize_json_results.py
to visualize json results.