Skip to content

Latest commit

 

History

History
33 lines (24 loc) · 1.37 KB

image_feature_extractor.md

File metadata and controls

33 lines (24 loc) · 1.37 KB
Raw

Info:

  • Available extractors: resnet-152
  • Features dimension: (1,2048) - without batch normalization, (1,256) - feature vector is linearly transformed to have the same dimension as the input dimension of the LSTM network
  • Required data: images (currently supports .jpg, .png)

Usage:

Invoke the extract_image_features.sh script the following way:

sbatch extract_image_features.sh environment image_folder_location output_folder_location resized_flag

Example:

sbatch extract_image_features.sh taito /proj/memad/COCO/resized_train2014/ ./features/ true

Details:

  • Rezizing of images will be skipped if the resized_flag passed is true
  • Images location is defaulted to COCO images based on the environment if the image_folder_location is not passed
  • Output features location is defaulted to ./features/ if the output_folder_location is not passed
  • Valid environments that can be passed for now are taito/triton
  • Features are saved as pickle (.pkl) files per image (image name is used as the name of the output file)

Other job trigger related details can be inferred from: https://version.aalto.fi/gitlab/CBIR/image_captioning/blob/image_feature_extractor/extract_image_features.sh

InProgress:

  • Making InceptionV3, AlexNET available
  • Unveiling options to customize extractors (different layers, dimensionality of the features)