Skip to content

lun-ai/Multi_tasks_Image_retrieval

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Multi-tasks-Image-retrieval

  • Caffe :/home/cuiyan/lun/docomo_hashing/DSH/
  • Script Code :/home/cuiyan/lun/docomo_hashing/DSH/scripts/docomo/
  • Docomo test Data :/home/cuiyan/ting/Data/Docomo/cloth_data/
  • Docomo train Data: /home/cuiyan/lun/docomo_train_image_set/

There is a brief outline of repository (README.txt) located under the path - /home/cuiyan/lun/docomo_hashing

1 Creating lmdb docomo training / testing data

  • /home/cuiyan/lun/docomo_hashing/DSH/scripts/docomo/

create_docomo_test_lmdb.sh Create test lmdb data from 273634 image data create_pattern_lmdb.sh Generate train/test/validation lmdb for pattern data create_color_lmdb.sh Generate train/test/validation lmdb for color data create_type_lmdb.sh Generate train/test/validation lmdb for type data make_mean.sh prepare binaryproto mean from image dataset

2 Train network

  • Please find train_.sh finetune_.sh under

/home/cuiyan/lun/docomo_hashing/DSH/TYPE/ /home/cuiyan/lun/docomo_hashing/DSH/PATTERN/ /home/cuiyan/lun/docomo_hashing/DSH/COLOR/

*Note: Logs and snapshots will be saved in each of the parent directories.

3 Extract binary features

  • /home/cuiyan/lun/docomo_hashing/DSH/scripts/docomo/

    extract_code.sh / extract_finetune_code.sh extract train data and labels binary code extract_code_docomo.sh extract 273634 test data binary code

4 Test network on partitions of data

  • /home/cuiyan/lun/docomo_hashing/DSH/scripts/docomo/

    test_map.m compute mean average precision on partition test set

5 Evaluate result of retrieval on docomo test dataset

  • /home/cuiyan/lun/docomo_hashing/DSH/scripts/docomo/

    docomo_prepare.m calculate hamming space distance from extract code docomo_get_retrieval.m sort 273634 image based on hamming space distance docomo_test_map.m evaluate mea average precision for top-N retrieval of 1000 queries

6 Repository Composition:

  • model_prototxt - a number of neural network model prototxts for future usage and documentation. These network includes simple self devised CNN, AlexNet, ResNet-50, VGG-16, ResNet-18 and GoogLenet.

  • model_weight - caffemodels that are likely to be reused.

About

Image retrieval based on deep neural network and deep supervised hashing methods

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published