About '--add-file-path' parameter

[LogSSim]

trafficstars

I ran the code but didn't understand what '--add-file-path' means. Could you explain this parameter?

[andvg3]

Hi @LogSSim ,

This parameter is used for handling the image resources, which is typically pointed to the previous Grasp-Anything version.

[Pierre0089]

Hi @andvg3,

Thanks for sharing so amazing datasets.

I am sorry I got lost in the use of this parameter. Do you mean file_path points to Grasp-Anything++ and add_file_path points to Grasp-anything?

Besides, I could not understand how to use "--dataset, --dataset-path, --add-file-path", e.g., $ python evaluate.py --dataset grasp-anywhere --dataset-path data/grasp-anything --add-file-path data/grasp-anything++/seen --iou-eval --seen 0 --use-depth 0 --network <path_to_pretrained_network> in Testing of ReadMe.

Hope for your reply, and appreciate for your help.

[andvg3]

Hi @Pierre0089 ,

Yes, just as the sample instruction in the readme file.

[Pierre0089]

Ok, thx.

What does "seen" mean in "--add-file-path"? Is it right to put all unzipped files from grasp_label_positive.zip in grasp-anything++ in data/grasp-anything++/seen?

[andvg3]

I think you should only put the seen images as in the .obj files, because, otherwise, the seen and unseen will likely overlap in the images. That may also cause the performance drop in https://github.com/Fsoft-AIC/LGD/issues/11#issuecomment-2440600036.

[Pierre0089]

I followed your advice, and modify control_manage_cornell.py to support the copy of grasp-anything++ dataset. After copying partial files of grasp_label_positive.zip and partial files of grasp_instructions, 1376 files are obtained, and put in "seen" directory, cf.

Unluckily, the grasp detection result is unchanged, still 0.28 instead of 0.37 in the paper.

The picture shows folders in grasp-anything++

Please correct me if I make any mistakes. Thx.

[LogSSim]

@andvg3 第一张是我的grasp_anything，第二张是我的grasp_anything_pp。我应该怎么调整呢

[andvg3]

I followed your advice, and modify control_manage_cornell.py to support the copy of grasp-anything++ dataset. After copying partial files of grasp_label_positive.zip and partial files of grasp_instructions, 1376 files are obtained, and put in "seen" directory, cf.

Unluckily, the grasp detection result is unchanged, still 0.28 instead of 0.37 in the paper.

The picture shows folders in grasp-anything++

Please correct me if I make any mistakes. Thx.

Hi, thanks for reporting. Maybe there were some problems with the number of samples in the splits. I’ll check it and do further updates if necessary.

[Pierre0089]

Thanks. Is there any problems with the number of samples in the splits?

[ZhaiTianLin]

Hi @andvg3,

Thanks for sharing so amazing datasets.

I am sorry I got lost in the use of this parameter. Do you mean file_path points to Grasp-Anything++ and add_file_path points to Grasp-anything?

Besides, I could not understand how to use "--dataset, --dataset-path, --add-file-path", e.g., $ python evaluate.py --dataset grasp-anywhere --dataset-path data/grasp-anything --add-file-path data/grasp-anything++/seen --iou-eval --seen 0 --use-depth 0 --network <path_to_pretrained_network> in Testing of ReadMe.

Hope for your reply, and appreciate for your help.

Hi @andvg3, 你好 ，

Thanks for sharing so amazing datasets.感谢您分享如此惊人的数据集。

I am sorry I got lost in the use of this parameter. Do you mean file_path points to Grasp-Anything++ and add_file_path points to Grasp-anything?很抱歉我在使用这个参数时迷失了方向。您的意思是 file_path 指向 Grasp-Anything++ 并且 add_file_path 指向 Grasp-anything 吗？

Besides, I could not understand how to use "--dataset, --dataset-path, --add-file-path", e.g., $ python evaluate.py --dataset grasp-anywhere --dataset-path data/grasp-anything --add-file-path data/grasp-anything++/seen --iou-eval --seen 0 --use-depth 0 --network <path_to_pretrained_network> in Testing of ReadMe.此外，我无法理解如何使用“--dataset，--dataset-path，--add-file-path”，例如 $ pythonvalu.py --dataset master-anywhere --dataset-path data/grasp自述文件测试中的 -anything --add-file-path data/grasp-anything++/seen --iou-eval --seen 0 --use-深度 0 --network <path_to_pretrained_network>。

Hope for your reply, and appreciate for your help.希望您的回复，并感谢您的帮助。

It's so confusing.

self.grasp_files = glob.glob(os.path.join(addition_file_path, 'positive_grasp', '*.pt'))
self.prompt_files = glob.glob(os.path.join(file_path, 'prompt', '*.pkl'))

Both of these should belong to Grasp++, with one being text prompt and the other being grasp data. However, their main paths are file_path and addition_file_path, which contradict each other.

[123liu172061]

Hello, I want to know if the mask is not used on the network. I look forward to your reply.