Skip to content
This repository has been archived by the owner on Oct 31, 2023. It is now read-only.

Add RetinaNet Implementation #102

Merged
merged 56 commits into from
Feb 15, 2019
Merged

Add RetinaNet Implementation #102

merged 56 commits into from
Feb 15, 2019

Commits on Oct 26, 2018

  1. Add RetinetNet parameters in cfg.

    Cheng-Yang Fu committed Oct 26, 2018
    Configuration menu
    Copy the full SHA
    e185c79 View commit details
    Browse the repository at this point in the history

  2. hot fix.

    Cheng-Yang Fu committed Oct 26, 2018
    Configuration menu
    Copy the full SHA
    6167fa4 View commit details
    Browse the repository at this point in the history

Commits on Oct 27, 2018

  1. Add the retinanet head module now.

    Cheng-Yang Fu committed Oct 27, 2018
    Configuration menu
    Copy the full SHA
    99920af View commit details
    Browse the repository at this point in the history

  2. Add the function to generate the anchors for RetinaNet.

    Cheng-Yang Fu committed Oct 27, 2018
    Configuration menu
    Copy the full SHA
    9e82436 View commit details
    Browse the repository at this point in the history

Commits on Oct 28, 2018

  1. Add the SigmoidFocalLoss cuda operator.

    Cheng-Yang Fu committed Oct 28, 2018
    Configuration menu
    Copy the full SHA
    69d5d3a View commit details
    Browse the repository at this point in the history

  2. Fix the bug in the extra layers.

    Cheng-Yang Fu committed Oct 28, 2018
    Configuration menu
    Copy the full SHA
    587ccd8 View commit details
    Browse the repository at this point in the history

  3. Change the normalizer for SigmoidFocalLoss

    Cheng-Yang Fu committed Oct 28, 2018
    Configuration menu
    Copy the full SHA
    89e35b2 View commit details
    Browse the repository at this point in the history

  4. Support multiscale in training.

    Cheng-Yang Fu committed Oct 28, 2018
    Configuration menu
    Copy the full SHA
    bd9a817 View commit details
    Browse the repository at this point in the history

Commits on Oct 29, 2018

  1. Add retinannet training script.

    Cheng-Yang Fu committed Oct 29, 2018
    Configuration menu
    Copy the full SHA
    882655a View commit details
    Browse the repository at this point in the history

Commits on Oct 31, 2018

  1. Add the inference part of RetinaNet.

    Cheng-Yang Fu committed Oct 31, 2018
    Configuration menu
    Copy the full SHA
    b5ca053 View commit details
    Browse the repository at this point in the history

Commits on Nov 2, 2018

  1. Fix the bug when building the extra layers in retinanet. 

    Update the matching part in retinanet_loss.
    Cheng-Yang Fu committed Nov 2, 2018
    Configuration menu
    Copy the full SHA
    a1f7365 View commit details
    Browse the repository at this point in the history

  2. Add the first version of the inference of RetinaNet. 

    Need to check it again to see if is there any room for speed
improvement.
    Cheng-Yang Fu committed Nov 2, 2018
    Configuration menu
    Copy the full SHA
    6cc7264 View commit details
    Browse the repository at this point in the history

  3. Remove the retinanet_R-50-FPN_2x.yaml first.

    Cheng-Yang Fu committed Nov 2, 2018
    Configuration menu
    Copy the full SHA
    dca2453 View commit details
    Browse the repository at this point in the history

Commits on Nov 3, 2018

  1. Optimize the retinanet postprocessing.

    Cheng-Yang Fu committed Nov 3, 2018
    Configuration menu
    Copy the full SHA
    ce06ecd View commit details
    Browse the repository at this point in the history

  2. Merge branch 'master' of https://github.com/facebookresearch/maskrcnn… 

    …-benchmark into retina_net
    Cheng-Yang Fu committed Nov 3, 2018
    Configuration menu
    Copy the full SHA
    615af53 View commit details
    Browse the repository at this point in the history

  3. quick fix.

    Cheng-Yang Fu committed Nov 3, 2018
    Configuration menu
    Copy the full SHA
    a859b1e View commit details
    Browse the repository at this point in the history

  4. Add script for training RetinaNet with ResNet101 backbone.

    Cheng-Yang Fu committed Nov 3, 2018
    Configuration menu
    Copy the full SHA
    2e9881f View commit details
    Browse the repository at this point in the history

Commits on Nov 6, 2018

  1. Move cfg.RETINANET to cfg.MODEL.RETINANET

    Cheng-Yang Fu committed Nov 6, 2018
    Configuration menu
    Copy the full SHA
    21a84e7 View commit details
    Browse the repository at this point in the history

  2. Remove the variables which are not used.

    Cheng-Yang Fu committed Nov 6, 2018
    Configuration menu
    Copy the full SHA
    ee7760f View commit details
    Browse the repository at this point in the history

  3. revert boxlist_ops. 

    Generate Empty BoxLists instead of [] in retinanet_infer
    Cheng-Yang Fu committed Nov 6, 2018
    Configuration menu
    Copy the full SHA
    adb25d6 View commit details
    Browse the repository at this point in the history

  4. Remove the not used commented lines. 

    Add NUM_DETECTIONS_PER_IMAGE
    Cheng-Yang Fu committed Nov 6, 2018
    Configuration menu
    Copy the full SHA
    b84ff0e View commit details
    Browse the repository at this point in the history

  5. remove the not used codes.

    Cheng-Yang Fu committed Nov 6, 2018
    Configuration menu
    Copy the full SHA
    b3af003 View commit details
    Browse the repository at this point in the history

  6. Move retinanet related files under Modeling/rpn/retinanet

    Cheng-Yang Fu committed Nov 6, 2018
    Configuration menu
    Copy the full SHA
    911196f View commit details
    Browse the repository at this point in the history

  7. Add retinanet_X_101_32x8d_FPN_1x.yaml script. 

    This model is not fully validated. I only trained it around 5000
iterations and everything is fine.
    Cheng-Yang Fu committed Nov 6, 2018
    Configuration menu
    Copy the full SHA
    5fc4b75 View commit details
    Browse the repository at this point in the history

Commits on Nov 10, 2018

  1. set RETINANET.PRE_NMS_TOP_N as 0 in level5 (p7), because previous set… 

    …ting may generate zero detections and could cause

the program break.
This part is used in original Detectron setting.
    Cheng-Yang Fu committed Nov 10, 2018
    Configuration menu
    Copy the full SHA
    c8c4bc7 View commit details
    Browse the repository at this point in the history

  2. Fix the rpn only bug when the training ends.

    Cheng-Yang Fu committed Nov 10, 2018
    Configuration menu
    Copy the full SHA
    cfe06d8 View commit details
    Browse the repository at this point in the history

Commits on Nov 19, 2018

  1. Merge branch 'master' of https://github.com/facebookresearch/maskrcnn… 

    …-benchmark into retinanet

Add the EmptyBox in retina/inference if there is no detection.
    Cheng-Yang Fu committed Nov 19, 2018
    Configuration menu
    Copy the full SHA
    bac17d6 View commit details
    Browse the repository at this point in the history

Commits on Nov 23, 2018

  1. Minor improvements

    @fmassa
    fmassa committed Nov 23, 2018
    Configuration menu
    Copy the full SHA
    9e2baa7 View commit details
    Browse the repository at this point in the history

  2. Comments and add Python-only implementation

    @fmassa
    fmassa committed Nov 23, 2018
    Configuration menu
    Copy the full SHA
    a8c919a View commit details
    Browse the repository at this point in the history

  3. Bugfix and remove commented code

    @fmassa
    fmassa committed Nov 23, 2018
    Configuration menu
    Copy the full SHA
    190e132 View commit details
    Browse the repository at this point in the history

Commits on Nov 25, 2018

  1. keep the generalized_rcnn same. 

    Move the build_retinanet inside build_rpn.
    Cheng-Yang Fu committed Nov 25, 2018
    Configuration menu
    Copy the full SHA
    8bab238 View commit details
    Browse the repository at this point in the history

Commits on Jan 23, 2019

  1. Merge branch 'master' of https://github.com/facebookresearch/maskrcnn… 

    …-benchmark into retinanet
    Cheng-Yang Fu committed Jan 23, 2019
    Configuration menu
    Copy the full SHA
    51bbb17 View commit details
    Browse the repository at this point in the history

Commits on Jan 25, 2019

  1. Add USE_C5 in the MODEL.RETINANET

    Cheng-Yang Fu committed Jan 25, 2019
    Configuration menu
    Copy the full SHA
    b328d22 View commit details
    Browse the repository at this point in the history

  2. Merge branch 'retinanet' of https://github.com/chengyangfu/maskrcnn-b… 

    …enchmark into retinanet
    Cheng-Yang Fu committed Jan 25, 2019
    Configuration menu
    Copy the full SHA
    0dac79d View commit details
    Browse the repository at this point in the history

  3. Add two configs using P5 to generate P6.

    Cheng-Yang Fu committed Jan 25, 2019
    Configuration menu
    Copy the full SHA
    328ea98 View commit details
    Browse the repository at this point in the history

Commits on Jan 28, 2019

  1. fix the bug when loading the Caffe2 ImageNet pretrained model.

    Cheng-Yang Fu committed Jan 28, 2019
    Configuration menu
    Copy the full SHA
    0177419 View commit details
    Browse the repository at this point in the history

Commits on Jan 29, 2019

  1. Reduce the code depulication of RPN loss and RetinaNet loss.

    Cheng-Yang Fu committed Jan 29, 2019
    Configuration menu
    Copy the full SHA
    128c491 View commit details
    Browse the repository at this point in the history

  2. Remove the comment which is not used.

    Cheng-Yang Fu committed Jan 29, 2019
    Configuration menu
    Copy the full SHA
    dc73a33 View commit details
    Browse the repository at this point in the history

Commits on Jan 30, 2019

  1. Remove the hard coded number of classes.

    Cheng-Yang Fu committed Jan 30, 2019
    Configuration menu
    Copy the full SHA
    a51abdc View commit details
    Browse the repository at this point in the history

Commits on Jan 31, 2019

  1. share the foward part of rpn inference.

    Cheng-Yang Fu committed Jan 31, 2019
    Configuration menu
    Copy the full SHA
    5c7b391 View commit details
    Browse the repository at this point in the history

  2. fix the bug in rpn inference.

    Cheng-Yang Fu committed Jan 31, 2019
    Configuration menu
    Copy the full SHA
    bec7cc1 View commit details
    Browse the repository at this point in the history

  3. Remove the conditional part in the inference.

    Cheng-Yang Fu committed Jan 31, 2019
    Configuration menu
    Copy the full SHA
    77e3626 View commit details
    Browse the repository at this point in the history

  4. Bug fix: add the utils file for permute and flatten of the box 

    prediction layers.
    Cheng-Yang Fu committed Jan 31, 2019
    Configuration menu
    Copy the full SHA
    a8ba755 View commit details
    Browse the repository at this point in the history

  5. Update the comment.

    Cheng-Yang Fu committed Jan 31, 2019
    Configuration menu
    Copy the full SHA
    dbbb6f9 View commit details
    Browse the repository at this point in the history

Commits on Feb 1, 2019

  1. quick fix. Adding import cat.

    Cheng-Yang Fu committed Feb 1, 2019
    Configuration menu
    Copy the full SHA
    997ae29 View commit details
    Browse the repository at this point in the history

Commits on Feb 3, 2019

  1. quick fix: forget including import.

    Cheng-Yang Fu committed Feb 3, 2019
    Configuration menu
    Copy the full SHA
    8c93b1d View commit details
    Browse the repository at this point in the history

Commits on Feb 4, 2019

  1. fix merge. 

    Merge branch 'retinanet' of https://github.com/chengyangfu/maskrcnn-benchmark into retinanet
    Cheng-Yang Fu committed Feb 4, 2019
    Configuration menu
    Copy the full SHA
    fb3fe10 View commit details
    Browse the repository at this point in the history

  2. Adjust the normalization part according to Detectron's setting.

    Cheng-Yang Fu committed Feb 4, 2019
    Configuration menu
    Copy the full SHA
    096f0d6 View commit details
    Browse the repository at this point in the history

  3. Pull from maskrcnn-benchmark master branch. 

    Merge https://github.com/facebookresearch/maskrcnn-benchmark into retinanet
    Cheng-Yang Fu committed Feb 4, 2019
    Configuration menu
    Copy the full SHA
    da19923 View commit details
    Browse the repository at this point in the history

Commits on Feb 13, 2019

  1. Use the bbox reg normalization term.

    Cheng-Yang Fu committed Feb 13, 2019
    Configuration menu
    Copy the full SHA
    afd5f0b View commit details
    Browse the repository at this point in the history

  2. Merge https://github.com/facebookresearch/maskrcnn-benchmark into ret… 

    …inanet
    Cheng-Yang Fu committed Feb 13, 2019
    Configuration menu
    Copy the full SHA
    c235307 View commit details
    Browse the repository at this point in the history

  3. Clean the code according to recent review.

    Cheng-Yang Fu committed Feb 13, 2019
    Configuration menu
    Copy the full SHA
    6389130 View commit details
    Browse the repository at this point in the history

Commits on Feb 14, 2019

  1. Using CUDA version for training now. And the python version for training 

    on cpu.
    Cheng-Yang Fu committed Feb 14, 2019
    Configuration menu
    Copy the full SHA
    26c707e View commit details
    Browse the repository at this point in the history

  2. rename the directory to retinanet.

    Cheng-Yang Fu committed Feb 14, 2019
    Configuration menu
    Copy the full SHA
    45372c4 View commit details
    Browse the repository at this point in the history

  3. Make the train and val datasets are consistent with mask r-cnn setting.

    Cheng-Yang Fu committed Feb 14, 2019
    Configuration menu
    Copy the full SHA
    3bb285d View commit details
    Browse the repository at this point in the history

  4. add comment.

    Cheng-Yang Fu committed Feb 14, 2019
    Configuration menu
    Copy the full SHA
    37e9075 View commit details
    Browse the repository at this point in the history