Skip to content

D
detectron
Switch branch/tag
  1. 31 Jan, 2018 5 commits
    • Ashwin Bharambe's avatar
      fbshipit-source-id: f79749dcfe9fc3ff0ecc2459f6c1d31e8cac8094 · 05d04d3a
      Ashwin Bharambe authored
      05d04d3a
    • Ashwin Bharambe's avatar
      Introduce DetectionModelHelper.GetLossScale() · 3e99b2f5
      Ashwin Bharambe authored 
      Summary:
Under a distributed setting (multiple nodes), the loss scale may need
to account for the number of nodes participating in the algorithm. This method
provides a reasonable way for providing this config.

Reviewed By: rbgirshick

Differential Revision: D6826976

fbshipit-source-id: d224200710b0f3c3245851874c9f4f3ab871710f
      3e99b2f5
    • Ashwin Bharambe's avatar
      Fix AffineChannel op param initialization · e59c30bb
      Ashwin Bharambe authored 
      Summary:
The `AffineChannel` op wrapper did not use the officially sanctioned
version of parameter creation, namely using `ModelHelper.create_param()`
Presumably this was because of the `share_with` option where you could share
parameters.

While doing this though, the parameters were never initialized! This probably
doesn't matter for the algorithm (those params get overwritten pretty quickly)
but it's a bug nevertheless. I decided to kill the `share_with` option since
this isn't used anyway and resort to the standard way of creating weights and
biases.

Reviewed By: rbgirshick

Differential Revision: D6826914

fbshipit-source-id: 65628c59b085b9ab160006b003dd40dbefa2f7c1
      e59c30bb
    • Ashwin Bharambe's avatar
      Refactor { train_net, utils/net } · 946ba8d0
      Ashwin Bharambe authored 
      Summary:
This diff starts a series of diffs to refactor the detectron codebase
so it can be used in an elastic data parallel context (see
https://our.intern.facebook.com/intern/dex/caffe2/elastic-data-parallel-model-for-distributed-training/)
without any forks.

Specifically, this diff does the following:

 - Splits out the `TrainingStats` class so it can be reused / composed
 - Slightly refactor of `initialize_weights_from_file()`
 - Reduces the verbosity of some logs

Hopefully, nothing controversial here :)

Reviewed By: rbgirshick

Differential Revision: D6826820

fbshipit-source-id: fc15209a0ff50e5d09281e36173198c77aa77a12
      946ba8d0
    • Ashwin Bharambe's avatar
      fbshipit-source-id: 1d8a1db27d816cf007ff52c1b031c18a177d638c · 8c013e44
      Ashwin Bharambe authored
      8c013e44
  3. 30 Jan, 2018 5 commits
    • Kaiming He's avatar
      improve print_net · 0bc0a8d4
      Kaiming He authored 
      Summary: Revise print_net to show the full forward model when there are multiple losses (which is always the case).

Reviewed By: rbgirshick

Differential Revision: D6829753

fbshipit-source-id: 53c1d63d95a78f83a0275b90cad1517315786e7d
      0bc0a8d4
    • Ross Girshick's avatar
      Add template for github issues · fa8dba9c
      Ross Girshick authored 
      Reviewed By: KaimingHe

Differential Revision: D6842986

fbshipit-source-id: c8370e2f831239b5489dc723738c6bdf5708a40d
      fa8dba9c
    • Peizhao Zhang's avatar
      Added a script to convert pkl model to pb. · 757d77c2
      Peizhao Zhang authored 
      Summary:
Added a script to convert pkl model to pb.
- Supported Faster R-CNN detection model.
- Supported converting to CPU and GPU model.
- Fused AffineChannel to Conv.
- Verified model after conversion.

Reviewed By: rbgirshick

Differential Revision: D6783633

fbshipit-source-id: 19706d3074a4a784a2161a695d7c534e014ebb3f
      757d77c2
    • Ilija Radosavovic's avatar
      Support running the demo with RetinaNet · dd6c6615
      Ilija Radosavovic authored 
      Reviewed By: rbgirshick

Differential Revision: D6829499

fbshipit-source-id: 66fe40ef1b2c8c560ce769faa27d107c3c7be841
      dd6c6615
    • Ilija Radosavovic's avatar
      Re-use test_engine roidb range testing logic for RetinaNet · b7159125
      Ilija Radosavovic authored 
      Reviewed By: rbgirshick

Differential Revision: D6829350

fbshipit-source-id: a5fbaa8900fab28602d7a993d63206ca67b24975
      b7159125
  5. 29 Jan, 2018 3 commits
  7. 24 Jan, 2018 3 commits
  9. 23 Jan, 2018 2 commits