Giters

switchablenorms / DeepFashion2

DeepFashion2 Dataset https://arxiv.org/pdf/1901.07973.pdf

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

The annotations does not correspond to paper

majercakdavid opened this issue · comments

commented

I have noticed, as well as read in here, that the full dataset was not release yet and the version currently available contains only half of all samples. That is why I have tried to plot the graphs, mentioned in the Figure 3, myself. I have observed misalignment when it comes to scale, I have not been able to check the other ones. The pie charts for different measurements are as follows:
scale
The first one is made from the annotations file so it just uses the pre-made scale labels. Second uses the bounding boxes/areas from the annotation files and the third one uses area of bounding boxes from segmentation masks created by pycocoutils package. The second and third compare the area of the bounding box to the area of the image and split the images into 3 buckets according to the ration with the ones <10% area, 10%-40% area and >40% area.

Which of them is correct approach? Why the details in annotations differ from calculated ones?

commented

We use bounding boxes/areas, which is the second one in your measurements, and the result is consistent with the first measurement, which uses the pre-made scale label.

commented

According to what I have seen in the dataset, these labels are to large extent wrong and it would be really awesome if you would consider re-assigning them. Examples of wrong bounding boxes can be seen in the following images:
bboxes_areas_1
bboxes_areas_2
bboxes_areas_3
bboxes_areas_4
The dashed lines are the ones from annotation files. The thick ones represent bounding boxes inferred from the segmentations.

commented

Could you please provide names of the above four images?

commented

There are many more similar to these, I could not find all of the previously mentioned ones but these are the ones I have found:

  • 089587.jpg
    bboxes_areas_089587

  • 080658.jpg
    bboxes_areas_080658

  • 005696.jpg
    bboxes_areas_3

  • 008506.jpg
    bboxes_areas_4

Moreover some of the segmentations are either missing or the items are split without a reason according to my visual assessment, e.g.:

  • 49276.jpg
    bboxes_segm_49276

  • 088005.jpg
    bboxes_segm_088005

Thank for your response and collaboration!

commented

Bounding boxes referred from segmentation label are accurate enough. You can use them instead.
A part of masks are annotated coarsely while the others are obtained through refinement.

commented

How we can we refer bounding boxes from segmentation label exactly.I am facing similar issue of wrong bounding box annotations