Commits · avg_reg_feats · Vision / GenVQA · GitLab

Snippets Groups Projects

Oct 13, 2016
- use averaged region features instead of averaging prob · 49a622c1
  tgupta6 authored 8 years ago
  
  49a622c1
Oct 12, 2016
- change params in constants_crunchy · 1a015d00
  tgupta6 authored 8 years ago
  
  1a015d00
- change mil loss to use scores instead of softmax prob · 582c88c7
  tgupta6 authored 8 years ago
  
  582c88c7
- add yes no features and detector scores as answering features · cf15ccfa
  tgupta6 authored 8 years ago
  
  cf15ccfa
Oct 11, 2016
- object detector scores as question answering features · aa661802
  tgupta6 authored 8 years ago
  
  aa661802
Oct 10, 2016
- constants_vision_gpu_2 · 2bb5fc27
  tgupta6 authored 8 years ago
  
  2bb5fc27
Oct 09, 2016
- constants_vision_gpu_1 update · 477b988e
  tgupta6 authored 8 years ago
  
  477b988e
- constants_crunchy that works with new obj loss/fine_tune with decayed lr · c8a4e586
  tgupta6 authored 8 years ago
  
  c8a4e586
- top k metrics for evaluation object detectors · 0ce3ba5e
  tgupta6 authored 8 years ago
  
  0ce3ba5e
- margin based loss for object training · 4d563c2d
  tgupta6 authored 8 years ago
  
  4d563c2d
- multi hot object label vector in cropped_regions_cached_features.py · 5a257b12
  tgupta6 authored 8 years ago
  
  5a257b12
- allow fine_tune to begin with decayed learning rate · 2216f3bc
  tgupta6 authored 8 years ago
  
  2216f3bc
- remove object and group from object_labels.json (modification in visual_genome_parser.py) · 726009a2
  tgupta6 authored 8 years ago
  
  726009a2
- changed genome test ids to simple · 69e525c6
  tgupta6 authored 8 years ago
  
  69e525c6
- added obj visualization flag in constants_crunchy · e4183eca
  tgupta6 authored 8 years ago
  
  e4183eca
Oct 08, 2016
- change to_region_with_hypernym_labels in constants_crunchy · 23179b1b
  tgupta6 authored 8 years ago
  
  23179b1b
- remove tailing : in losses · 696cacfb
  tgupta6 authored 8 years ago
  
  696cacfb
- changes object loss in answer train script to measure presence in top 5 · 608abd82
  tgupta6 authored 8 years ago
  
  608abd82
- added metrics for object classification performance · f290b6ef
  tgupta6 authored 8 years ago
  
  f290b6ef
- no fractional object prob in gt label vector · a0c4b8aa
  tgupta6 authored 8 years ago
  
  a0c4b8aa
- multilabel_margin_loss · f6fa768f
  tgupta6 authored 8 years ago
  
  f6fa768f
- if unknown then no gradient (changed object labels in cropped_regions_cached_features.py) · 98f2b8b5
  tgupta6 authored 8 years ago
  
  98f2b8b5
- add_hypernyms fun in visual_genome_parser · f88c7cd9
  tgupta6 authored 8 years ago
  
  f88c7cd9
- flag for visualizing object prediction · 7e43c711
  tgupta6 authored 8 years ago
  
  7e43c711
Oct 06, 2016
- removed arg space_dim from answer_classifier_cached_features/train.py and inference.py · 708f8389
  tgupta6 authored 8 years ago
  
  708f8389
- simple genome split and cleaner constants file · 7fc938b4
  tgupta6 authored 8 years ago
  
  7fc938b4
- make ans eval and train compatible with each other (resolved conflict in ans train) · 2ed2f60a
  tgupta6 authored 8 years ago
  
  Conflicts: answer_classifier_cached_features/train.py
  2ed2f60a
- obj atr eval script now can write html files to display results · 89f48388
  tgupta6 authored 8 years ago
  
  89f48388
Sep 30, 2016
- l2 on word vectors and put back relu and bn in obj atr network · ab98b61f
  tgupta6 authored 8 years ago
  
  ab98b61f
Sep 28, 2016
- global step in optimizer · e4728984
  tgupta6 authored 8 years ago
  
  e4728984
- constants gpu1 change · ef44688d
  tgupta6 authored 8 years ago
  
  ef44688d
- obj atr with mil and ans loss working and also lr decay · 3758e7a5
  tgupta6 authored 8 years ago
  
  3758e7a5
- bn in answer inference · 849c57e3
  tgupta6 authored 8 years ago
  
  849c57e3
- scratch initializer in train script and mil loss · 7f64b299
  tgupta6 authored 8 years ago
  
  7f64b299
- vector encoding of nouns and attributes for mil loss · ab5bacd1
  tgupta6 authored 8 years ago
  
  ab5bacd1
Sep 27, 2016
- add missing self. · 1d8b3321
  tgupta6 authored 8 years ago
  
  1d8b3321
- add batchnorm in final answer scoring net and 2 extra features · b1c00c13
  tgupta6 authored 8 years ago
  
  b1c00c13
Sep 21, 2016
- inner_product_selection functional but not a good idea · e6318fb6
  tgupta6 authored 8 years ago
  
  e6318fb6
Sep 19, 2016
- wrap_elementwise_multiply · 97d8c1ed
  tgupta6 authored 8 years ago
  
  97d8c1ed
Sep 14, 2016
- relevance visualization · fed5bca0
  tgupta6 authored 8 years ago
  
  fed5bca0