search

Gaoyiminggithub / Graphonomy

Graphonomy: Universal Human Parsing via Graph Transfer Learning
MIT License
295 stars 68 forks source link

How to distinguish instances #3

Closed JonathanLehner closed 5 years ago

JonathanLehner commented 5 years ago

How can we get the instances? If there are two humans in the image, can we get a list of the areas that belong to each of them?

Gaoyiminggithub commented 5 years ago

Hi, If you want to get the pred of instances, you can get the instances by using the pretrained Mask-rcnn to detect the instance human mask.

JonathanLehner commented 5 years ago

great, thank you! I will try it

JonathanLehner commented 5 years ago

where is the model?

Gaoyiminggithub commented 5 years ago

You can get the pretrained Mask-rcnn model in here. ( I have used this pretrained model to get the instance mask, and then combined with our parsing results that we could get the stoa performance of instance human parsing.)

JonathanLehner commented 5 years ago

Thanks! how did you combine the results?

Gaoyiminggithub commented 5 years ago

We use the instance person mask to detect the instance part of the human part. (e.g., we have a human instance mask and a semantic segmentation, we use the instance person mask to detect the person part that belongs to which instance human.)