Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Oct 3, 2017

• Suppress unk when beam search.

  ruotianluo

  committed
  13516df

Commits on Sep 26, 2017

• Make a common father class CaptionModel (the original CaptionModel is renamed as OldModel). The new CaptionModel is designed to reduce the code used for beam search.

  Show description for 35185f5
  ruotianluo

  committed
  35185f5

• Stop expand finished beam when beam search.

  ruotianluo

  committed
  cca154d

Commits on Sep 5, 2017

• Fix the final bug in CaptionModel sample_beam.

  ruotianluo

  authored
  097ed64

• Again further fix bugs in CaptionModel in sample_beam

  ruotianluo

  authored
  35719f3

• Futher fix the sample_beam in CaptionModel.

  ruotianluo

  authored
  4880150

• Fix sample_beam in CaptionModel.

  ruotianluo

  authored
  c16f982

Commits on Sep 4, 2017

• Make sure to use vocab in infos when evaluating.

  ruotianluo

  committed
  0694263

• Fix the problem of duplicating name in dataloaderaw.

  ruotianluo

  committed
  a88a6d8

• Fix that the pretrained model doens't have use_att option.

  ruotianluo

  committed
  0aeb222

Commits on Aug 31, 2017

• Use torch.utils.data to load file.

  ruotianluo

  committed
  b58e41d

Commits on Aug 11, 2017

• Disable scheduled sampling during evaluation mode.

  ruotianluo

  committed
  0b8c057

• Fix a bug in resnet_utils.

  ruotianluo

  committed
  954b2ec

Commits on Aug 7, 2017

• Don't load attention features if the model is not attention model.

  ruotianluo

  committed
  0155760

Commits on Aug 5, 2017

• Remove coco/.

  ruotianluo

  committed
  b722828

• Rewrite the readme.

  ruotianluo

  committed
  8024c33

• Change the name prepro_feats_npy to prepro_feats

  ruotianluo

  committed
  fdf9fde

• Add option in eval.py to use different resnets. (Must match with what you use to preprocess.)

  ruotianluo

  committed
  062118e

Commits on Aug 2, 2017

• Specify the location of pretrained resnet model.

  ruotianluo

  committed
  879971b

• Save the histories of loss lr etc. in a histories pkl file to reduce the original infos file, because infos are needed for evaluation but histories are actually not necessary.

  ruotianluo

  committed
  2eb03b6

• Allow both with tensorflow and without tensorflow to remove the redundant code (train_tb.py).

  ruotianluo

  committed
  e787220

• Move all models from misc to models folder.

  ruotianluo

  committed
  e4d4811

• 1. Add Top Down attention model, it achieves the best performance now.

  Show description for 7f4f268
  ruotianluo

  committed
  7f4f268

• Fix dropout problems:

  Show description for 4ec0819
  ruotianluo

  committed
  4ec0819

Commits on Aug 1, 2017

• Change mean because of 0.2 behaviour.

  ruotianluo

  committed
  93ddeb4

Commits on Jul 18, 2017

• Use inferred shape to expand. (related to last commit.)

  ruotianluo

  committed
  9d66010

• Update AttModel.py

  

  raoyongming

  authored and

  ruotianluo

  committed
  bdc4a97

Commits on Jul 6, 2017

• Add pretrained link.

  Show description for 72c35dd
  ruotianluo

  committed
  72c35dd

Commits on Jun 28, 2017

• Forgot bmm for show attend tell.

  ruotianluo

  committed
  1f6f77b

• Use bmm to get attentioned feature. (faster)

  ruotianluo

  committed
  68ae965

Commits on Jun 27, 2017

• Remove repackage in utils.py

  ruotianluo

  committed
  48c4d53

Commits on Jun 26, 2017

• fix bug for dump image

  ruotianluo

  committed
  3a4ab11

Commits on Jun 21, 2017

• 1. Allow prepro_feats_npy to specify model root

  Show description for ca9111d
  ruotianluo

  committed
  ca9111d

Commits on Jun 20, 2017

• Merge branch 'master' into py3

  Show description for 141879d
  ruotianluo

  committed
  141879d

• Fix a typo in prepro_feats_npy.

  ruotianluo

  committed
  052b647

Pagination

PreviousNext