GitHub - ankitnayan/bundle_min_hashing_source: Implementation of Bundle Min Hashing for Logo Recognition

ankitnayan / bundle_min_hashing_source Public

Notifications You must be signed in to change notification settings
Fork 5
Star 4

Implementation of Bundle Min Hashing for Logo Recognition

Notifications

Name		Name	Last commit message	Last commit date
Latest commit History 2 Commits
match_sketch		match_sketch
rootSift		rootSift
.gitignore		.gitignore
count_keys.py		count_keys.py
create_h5.py		create_h5.py
create_keys.py		create_keys.py
create_sketch.cpp		create_sketch.cpp
fastcluster.py		fastcluster.py
h52txt.py		h52txt.py
jpg2pgm.cpp		jpg2pgm.cpp
load_sketches_redis_selectdb.cpp		load_sketches_redis_selectdb.cpp
readme		readme
results_classified_sketches_logosonly.csv		results_classified_sketches_logosonly.csv
sketch.h		sketch.h
sketches_train_list.txt		sketches_train_list.txt

Repository files navigation

Vocab Size: 1,000,003 (nearest prime number to 1M to keep hashing simple)
Num Bundes: 5,288,778
Num Keypoints: 11,888,319


Please Note
1. Implementation includes only till results by collision count. No tf-idf, cosine similarity or further implementations
2. Precision of 97.6% and Recall of 80.42% was obtained on Flickr-32 dataset with 4280 training images and 960 query images
3. All timings are unoptimized


Results - added file "results_classified_sketches_logosonly.csv" containing the classification result of each test image

Documentation - https://docs.google.com/document/d/1QAtGysfp57SebSXBPePj1FwiI0TtvqHxo7ZMOkStQYs/edit

Dictionary - https://drive.google.com/file/d/1RJvNRtawP70g3xuKR-f5Ka7VbCuk02uN/view

RedisDB  - https://drive.google.com/file/d/1Or9cxDDsh51u6AbdJcQ6uO_ffMT_ql8_/view