search

SHI-Labs / OneFormer

OneFormer: One Transformer to Rule Universal Image Segmentation, arxiv 2022 / CVPR 2023
https://praeclarumjj3.github.io/oneformer
MIT License
1.41k stars 128 forks source link

Mapillary dataset not working in google colab #28

Closed Raunak-ghost closed 1 year ago

Raunak-ghost commented 1 year ago

Hi so I tried to clone the repo in your google colab sample and modified the code to include mapillary it seems to be failling with a meta dictonary erorr in get_config where meta info seems to missing i.e thingstocontiguous id

praeclarumjj3 commented 1 year ago

Hi @Raunak-ghost, thanks for your interest in our work. Could you please share the complete Traceback error log?

Please ensure you are using the correct dataset registry: mapillary_vistas_panoptic_train, as the things_to_contiguous_id key does exist for the Mapillary Vistas Panoptic dataset.

https://github.com/SHI-Labs/OneFormer/blob/761189909f392a110a4ead574d85ed3a17fbc8a7/oneformer/data/datasets/register_mapillary_vistas_panoptic.py#L483

Raunak-ghost commented 1 year ago

Yes, so I just completely cloned the current version of the directory on Google Colab and it's crashing. Also, I can't use the weights and config in detectron2. It gives me a yaml error. How do I use it with detectron2 ? Also installed it on a local system and it seems to be giving the same yaml error.Attaching the screenshot. Could you please update the google colab example mentioned in the repository https://github.com/SHI-Labs/OneFormer There seems to be a duplicate repository as well which is actually being used in the google colab example which is - https://github.com/SHI-Labs/OneFormer-Colab

The google colab example used in the former github repo for which I have created the issue is working on this github repo and when I try to use the current one it crashes with different errors I am trying to replicate the issue on google collab will send you a link as soon as I do.

I am a masters student at Dalhousie University Canada trying to do a project for object detection.

Regards, Raunak Bhargava

On Wed, Mar 8, 2023, 3:09 p.m. Jitesh Jain @.***> wrote:

Hi @Raunak-ghost https://github.com/Raunak-ghost, thanks for your interest in our work. Could you please share the complete Traceback error log?

Please ensure you are using the correct dataset registry: mapillary_vistas_panoptic_train, as the things_to_contiguous_id key does exist for the Mapillary Vistas Panoptic dataset.

https://github.com/SHI-Labs/OneFormer/blob/761189909f392a110a4ead574d85ed3a17fbc8a7/oneformer/data/datasets/register_mapillary_vistas_panoptic.py#L483

— Reply to this email directly, view it on GitHub https://github.com/SHI-Labs/OneFormer/issues/28#issuecomment-1459874579, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFZAQPPYY2UYQNW3ZD3OOATW3BHWFANCNFSM6AAAAAAVTMTPDM . You are receiving this because you were mentioned.Message ID: @.***>

praeclarumjj3 commented 1 year ago

Also installed it on a local system and it seems to be giving the same yaml error. Attaching the screenshot.

Hi @Raunak-ghost, Sorry, I do not see any screenshots/error information in the previous comment.

Raunak-ghost commented 1 year ago

Hi, I Forgot to attach the screenshot but I tried to use the mapillary weights. The meta data issue gets resolved but the model seems to be taking a lot of memory to be run whether locally on a gpu or even a cpu both are crashing.Expect the same on Google colab as well I couldn't run a make on google collab which results in a lot of dependency errors, if you could update the google coab library to reflect that.

How do I use the mapillary pretrained weights is my question. I have 2 3090 Gpu which does not seem to be enough memory for the model although it does get initiated its ability to run predictions before running out of memory not to mention multi Gpu does not work. Regards, Raunak Bhargava

praeclarumjj3 commented 1 year ago

as well I couldn't run a make on google collab which results in a lot of dependency errors, if you could update the google coab library to reflect that.

Hi @Raunak-ghost, as I asked earlier, please share the errors. I can't help you with a vague lot of dependency errors description.

For the memory issue, are you able to run ADE20K/Cityscapes/COCO models? If yes, there's probably something else at play here.

praeclarumjj3 commented 1 year ago

Closing due to inactivity, feel free to open if you have any more issues.