Broken URLs about fashion-iq

[yan9qu]

This page may help to solve this issue. https://github.com/lugiavn/notes/blob/master/fashioniq_tirg.md

[pisnitz]

I met the same problem. Could anybody figure this out?

[yeeeqichen]

Hi

I finally solve this by manually remove the broken images from the split file.

You could find the broken images with following code:

from PIL import Image
import os

prefix = 'The Prefix To The Data File, You Need To Customize This'
with open('broken_img.txt', 'w') as f1:
    for file in os.listdir(prefix + 'dress/'):
        with open(prefix + 'dress/' + file, 'rb') as f:
            try:
                img = Image.open(f).convert('RGB')
            except OSError:
                print('broken', file)
                f1.write(file + '\n')

[pisnitz]

Hi

I finally solve this by manually remove the broken images from the split file.

You could find the broken images with following code:

from PIL import Image
import os

prefix = 'The Prefix To The Data File, You Need To Customize This'
with open('broken_img.txt', 'w') as f1:
    for file in os.listdir(prefix + 'dress/'):
        with open(prefix + 'dress/' + file, 'rb') as f:
            try:
                img = Image.open(f).convert('RGB')
            except OSError:
                print('broken', file)
                f1.write(file + '\n')

Thanks a lot! But I got another question. If you mannully remove the broken images from the split file, how to match up the candidate and target in caption file? Removing the caption once the broken image appears in candidate list or target list?

[yan9qu]

Hi

I finally solve this by manually remove the broken images from the split file.

You could find the broken images with following code:

from PIL import Image
import os

prefix = 'The Prefix To The Data File, You Need To Customize This'
with open('broken_img.txt', 'w') as f1:
    for file in os.listdir(prefix + 'dress/'):
        with open(prefix + 'dress/' + file, 'rb') as f:
            try:
                img = Image.open(f).convert('RGB')
            except OSError:
                print('broken', file)
                f1.write(file + '\n')

Thank you. And may I ask can you get all the images?

[yeeeqichen]

Hi
I finally solve this by manually remove the broken images from the split file.
You could find the broken images with following code:

from PIL import Image
import os

prefix = 'The Prefix To The Data File, You Need To Customize This'
with open('broken_img.txt', 'w') as f1:
    for file in os.listdir(prefix + 'dress/'):
        with open(prefix + 'dress/' + file, 'rb') as f:
            try:
                img = Image.open(f).convert('RGB')
            except OSError:
                print('broken', file)
                f1.write(file + '\n')

Thanks a lot! But I got another question. If you mannully remove the broken images from the split file, how to match up the candidate and target in caption file? Removing the caption once the broken image appears in candidate list or target list?

Yep :)

[yeeeqichen]

Hi
I finally solve this by manually remove the broken images from the split file.
You could find the broken images with following code:

from PIL import Image
import os

prefix = 'The Prefix To The Data File, You Need To Customize This'
with open('broken_img.txt', 'w') as f1:
    for file in os.listdir(prefix + 'dress/'):
        with open(prefix + 'dress/' + file, 'rb') as f:
            try:
                img = Image.open(f).convert('RGB')
            except OSError:
                print('broken', file)
                f1.write(file + '\n')

Thank you. And may I ask can you get all the images?

Sorry, I couldn't get all the images neither. I just remove the broken images from the dataset and run the code with the modified dataset, which might lead to different performance compared to the results reported in related paper.