Trouble using mhp_extension

[jeremy-computervision]

trafficstars

Hi,

Thank you for your interesting work. I've been trying to use the demo.ipynb for multi human parsing but it seems that I need to create the demo_val.json when calling the function register_coco_instances. Could you explain how to use this for my own images ?

Thanks for your help,

Jeremy

[hazel0504]

Any solution to this?

[y-zheng18]

I've got the same problem. Did you solve it?

[melih-unsal]

Any solution?

[GoGoDuck912]

@XuYunqiu Yunqiu, Can you help to look at this issue?

[Julymycin]

@melih1996 you can create a presudo coco instance label file (.json) for your own images, I think. I used mhp for inference two months ago, but I can't remember very clearly now, sorry. Maybe you can refer to mask2coco modified from human_to_coco.

[XuYunqiu]

@melih1996 you can create a presudo coco instance label file (.json) for your own images, I think. I used mhp for inference two months ago, but I can't remember very clearly now, sorry. Maybe you can refer to mask2coco modified from human_to_coco.

@Julymycin is right. For the custom datasets, data should be converted to COCO-style, and the custom datasets should be registered. The codes for creating COCO-style annotation can be found in this folder. And the custom datasets with COCO-style annotations can be registered like following few lines https://github.com/PeikeLi/Self-Correction-Human-Parsing/blob/ebe6cebd673aa7b9373cbcf4191a0b2c965dfd70/mhp_extension/detectron2/tools/finetune_net.py#L49-L52

[dvschultz]

Can you update the demo.ipynb to include these steps?