Zero-shot COCO instance segmentation training and evaluate

[hhaAndroid]

Hello developers, I want to perform zero-shot instance segmentation on COCO based on this repository(I mainly want to verify the reliability of the automatic labeling process), and evaluate the difference between the generated annotations and the ground truth annotations. Finally, I want to quickly verify the annotation quality using Mask R-CNN in mmdetection. Are you currently working on this process? I can submit code in the form of a pull request.

[rentainhe]

Hello developers, I want to perform zero-shot instance segmentation on COCO based on this repository(I mainly want to verify the reliability of the automatic labeling process), and evaluate the difference between the generated annotations and the ground truth annotations. Finally, I want to quickly verify the annotation quality using Mask R-CNN in mmdetection. Are you currently working on this process? I can submit code in the form of a pull request.

We really appreciate the PR from the community users~ As you can see, it's just a simple demo by combining GroundingDINO and Segment-Anything now~

We are also curious about the zero-shot instance seg results on coco by grounded-segment-anything~ We will try to test the zero-shot COCO segmentation these days~ @hhaAndroid , if you have already tested some results~ we will really appreciate it~

[hhaAndroid]

Hello developers, I want to perform zero-shot instance segmentation on COCO based on this repository(I mainly want to verify the reliability of the automatic labeling process), and evaluate the difference between the generated annotations and the ground truth annotations. Finally, I want to quickly verify the annotation quality using Mask R-CNN in mmdetection. Are you currently working on this process? I can submit code in the form of a pull request.

We really appreciate the PR from the community users~ As you can see, it's just a simple demo by combining GroundingDINO and Segment-Anything now~

We are also curious about the zero-shot instance seg results on coco by grounded-segment-anything~ We will try to test the zero-shot COCO segmentation these days~ @hhaAndroid , if you have already tested some results~ we will really appreciate it~

I'll try it out in the next two days. If there are results I will give you feedback.

[rentainhe]

huanghaian

Sure!

[rentainhe]

@rentainhe Do you have personal WeChat that we can add? I think it's better to stay in touch through WeChat. You can add my WeChat: huanghaian.

I'm not sure if I added the correct WeChat ID. My WeChat ID is rentianhe666. If it's convenient for you, you can add me anytime.

[nemonameless]

Have you reproduced the mask AP of COCO or LVIS as Table 5?

[hhaAndroid]

I am unable to reproduce the zero-shot accuracy of the grounding dino and need some help

[Abhaykumar04]

.