The pretrained resnet152 from torchvision doesn't work about arel HOT 5 CLOSED

Please see https://github.com/eric-xw/AREL/tree/master/misc for more details of the resnet model.

from arel.

It's my fault to miss the extract features script, thanks for your time!

from arel.

Sorry to bother you again, i still don't know how to resize the original image to (256, 256) as you did in "/mnt/sshd/wenhuchen/VIST/images_256/{}/". I chose the image of 800 * 600 from the json file such as train.story-in-sequence.json, but i can't get the same fc_features as the loading of .npy file when i try the resize method in cv or Image library.

from arel.

This is the script we used to extract the ResNet features from images: https://github.com/eric-xw/AREL/blob/master/scripts/extract_features.py

from arel.

Hi, despite using this script the results for me varied , was there some pre-processing (resizing etc ) to the original image before doing this step ?
That is steps to get the resnet image features from say this raw image which is part of VIST -> https://ibb.co/qxjgTtZ
Would appreciate all inputs.

from arel.