cvdfoundation / fashionpedia Goto Github PK

I found this dataset is similar with kaggle competition imaterialist-fashion-2020-fgvc7. The pictures of the two datasets are similar, but the resolution is different, and the label format is different. Can you elaborate on the relationship between these two data sets？

Hello, happy new year, and thank you for this work!

I was wondering if you had any results for semantic segmentation on this dataset or are aware of any works that have such results?

I appreciate your time.

Thank you!
Josh

I downloaded the instances_attributes_train2020.json and instances_attributes_val2020.json. And I found weird segmentation format . You can search "counts" in the file. Such as these rows:

"segmentation": {
"size": [
1024,
688
],
"counts": "emX8;do02M3M4K4L4L3M4M2M3M4L3M4L3N2B?L3M3]LAhVO[NKU2eh00UWOaM3o1\\g0S2WXOUL7j1\\g0j51O001O1O1O001O1N2O0O2O001N101O0O2O0O1O1O1O1O1O1YOgWOfIZh0U6oWOfIRh0X6UXObIlg0]6[XO\\Ifg0b6m0O1N2K5J6\\O]VOeJii0V5c0M3[ObUOdKbj0Z4c0N2N2M3N2J6H8H8I7E;E;E;E;E<G8J6K5J7I7I0C[Yh:"
},

Did the file have a problem? Hoping for help.

Great dataset !

I have some questioins regarding the attribute annotations. I'm wondering if the attributes is related to the categories. For example, 'classic military (jacket)' should be a subcategory for 'jacket'. But there are attributes like 'hoodie', 'blazer' which don't have any information about how they are related to the categories. This makes it hard to filter the attribute prediction with the classification result of the category and determine which categories are mutual exclusive.

Also, it's there a way to get a complete version of this image.

Any help to solve the above problem will be appreciated.

Hi, so i have been playing with instance segmentation for the past week and cam around this library called Pixel lib

https://pixellib.readthedocs.io/en/latest/Image_instance.html

which is amazing at Instance Segmentation with few lines of Code.

Now it works on People,animals but has a problem on Fashion (Especially Shoes Dataset)

So, I Finally found the Fashion Pedia dataset and annotations but how can i use it with Pixel lib ?

My Code on Kaggle:

pip install pixellib

import pixellib

from pixellib.instance import instance_segmentation  

from matplotlib import pyplot as plt


!wget "https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5"


segmentation_model = instance_segmentation()

segmentation_model.load_model("./mask_rcnn_coco.h5")

segmask,output = segmentation_model.segmentImage("../input/facess/8.jpg", extract_segmented_objects = True, save_extracted_objects = True, show_bboxes = True, output_image_name = "output1.jpg")

plt.imshow(Image.open("./segmented_object_1.jpg"))

https://i.postimg.cc/BbnzTW4j/Screenshot-2021-10-26-at-18-53-53-notebook7bd3db2b2e-Kaggle.png

The below image i want to do image segmentation ? how can i do that ?

https://i.postimg.cc/bNGCp9Jf/116.jpg

Thanks

Hi, everyone.. so i have many pics of footwear.

i want to do instance segmentation of my footwear and remove the background.

how can use fashionpedias API with my custom dateset ? or do i have to annotate them one by one

Annotation for "attributes" in "instances_attributes_val2020" json looks something like,
'attributes': [{'id': 234, 'name': 'no opening', 'supercategory': 'opening type', 'level': 1, 'taxonomy_id': 'att000242_00'}, {'id': 281, 'name': 'plastic', 'supercategory': 'non-textile material type', 'level': 1, 'taxonomy_id': 'att000298_00'},........, {'id': 340, 'name': 'plant', 'supercategory': 'textile pattern', 'level': 1, 'taxonomy_id': 'att000363_00'}]
there is a jump in ID from 234 to 281 and it ends on 340.
I am assuming there are some wrong annotations, correct me if I am wrong.
if these annotations are correct, then how one can inference attribute from category? where to get relation between category and attribute?
In case of inferencing done through fashionpedia detection, How one can extract attribute values from the saved check point?

Hi, thanks for your great work.

I found there are many gray-scale images in the training set. Is this expected?

like ac7f353fd7481926e8fa9eafed609948.jpg, 7d956f0919f2eab63e481e9321ce2028.jpg and so on