Segmentation mask overlapping

[ttacfq]

Hi,

Great appreciate for this work! I am trying to use this dataset to do semantic segmentation and I found the segmentation annotations overlap within different categories, so the overlapping part will not belong to only one class, which may lead to a multi-label classification.

Could you please share how you use the dataset to train a segmentation model and which model you have used?

Thanks!

[geyuying]

First, a bounding box is detected and each pixel within this bounding box will be assigned a mask label (For each detected bounding box, it will output knn, where k is the number of clothes categories and n is the size of feature map). For details, you can refer to Mask R-CNN.

We use the model below for clothes segmentation: https://github.com/facebookresearch/Detectron/blob/master/configs/12_2017_baselines/e2e_mask_rcnn_R-50-FPN_1x.yaml.

[ttacfq]

Thanks for your reply! So you are using maks-rcnn to get the bounding box first and then get the semantic segmentation (contour) mask? ! was wondering if I don't use the bounding box and only use the segmentation annotations, how to generate ground truth to train a segmentation model, because there are some overlapping around the waist (some pixels belong to both skirt and outwear). Are you using layered binary segmentation mask for each clothing category?

[geyuying]

In DeepFashion2, each clothing item is annotated with a bounding box and a polygon within the bounding box. Binary segmentation mask can be generated from polygon with cocoapi.

[HansolEom]

In DeepFashion2, each clothing item is annotated with a bounding box and a polygon within the bounding box. Binary segmentation mask can be generated from polygon with cocoapi.

Could you please release your code? I want deepfashion2's segmentation binary.