mirrors
/
fast-reid
mirror of https://github.com/JDAI-CV/fast-reid.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity
212 Commits
1 Branch
4 Tags
15 MiB
Commit Graph

90 Commits (eebc71b7c8ac13f6b84a7ccc1b311c902af14aec)

Author SHA1 Message Date
liaoxingyu ecc2b1a790 update naive sampler 
Summary: update naive sampler which will introduce unbalanced sampling
 2020-06-15 20:50:25 +08:00
liaoxingyu 56a1ab4a5d update fast global avgpool 
Summary: update fast pool according to https://arxiv.org/pdf/2003.13630.pdf
 2020-06-12 16:34:03 +08:00
liaoxingyu cbdc01a1c3 update pairwise circle loss 
Summary: add param of pairwise circle loss to config, and update pairwise circle loss version
 2020-06-10 19:07:29 +08:00
liaoxingyu 96b9ad2d99 fix docs in ToTensor 2020-06-10 19:06:03 +08:00
liaoxingyu 3732f94405 update osnet 2020-06-09 14:38:49 +08:00
liaoxingyu 25a7f82df7 change style in baseline 2020-06-05 11:23:11 +08:00
liaoxingyu bc221cb05f fix mgn multi-gpu training problem 
Summary: norm_type in pool_reduce will not change when use syncBN
 2020-06-05 11:11:50 +08:00
liaoxingyu 94d85fe11c fix convert caffe model problem 2020-06-04 16:39:12 +08:00
liaoxingyu e7156e1cfa fix mgn not registered problem 2020-06-03 11:46:28 +08:00
liaoxingyu 85672b1e75 add circle & arcface layer info 
Summary: show num_features and num_classes in circle & arcface layer, like nn.Linear
 2020-05-31 15:50:56 +08:00
liaoxingyu e2761f7581 fix style in comm dataset 2020-05-30 16:56:08 +08:00
liaoxingyu c036ac5bdd update reduction head 2020-05-30 16:50:02 +08:00
liaoxingyu 36c04f0a9f fix resume training problem 
Summary: when resume training, need to reiter dataloader because we update pid_dict in dataset, but dataloader with multiprocess won't do the same update
 2020-05-30 16:44:18 +08:00
liaoxingyu 5528d17ace refactor code 
Summary: change code style and refactor code, add avgmax pooling layer in gem_pool
 2020-05-28 13:49:39 +08:00
liaoxingyu cd69ba6ffb update augmix augments 
Summary: replace augmix augmentations for adapting reid
 2020-05-27 22:56:31 +08:00
liaoxingyu a1cb123cfa fix R101 bottleneck missing problem 
Summary: add key 101 in block dict to support R101
 2020-05-26 14:48:32 +08:00
liaoxingyu d4b71de3aa switch between soft and hard margin when inf 
Summary: Add a mechnism to automatic switch triplet loss with soft margin to hard margin when loss becomes inf.
 2020-05-26 14:36:33 +08:00
liaoxingyu 5982f90920 support loading various pretrained weights 
Summary: Support loading pretrained model by custom path. With this function, we can load infoMin weights.
 2020-05-26 14:33:18 +08:00
liaoxingyu 5d4758125d support ResNet34 backbone 
Summary: add BasicBlock to support ResNet34
 2020-05-26 13:18:09 +08:00
liaoxingyu 84c733fa85 fix: remove prefetcher, put normalizer in model 
1. remove messy data prefetcher which will cause  confusion
2. put normliazer in model to accelerate training via GPU computing
 2020-05-25 23:39:11 +08:00
liaoxingyu 94c86579a3 fix(heads): fix bug in reduce head 
add neck_feat from config, add inplace in leakyrelu for memory save
 2020-05-23 10:41:13 +08:00
liaoxingyu c21de64166 fix: add linear initial method 2020-05-21 23:59:51 +08:00
liaoxingyu e990cf3e34 style: fix some typro 2020-05-21 15:55:51 +08:00
liaoxingyu 2ac55a7601 feat: update roc curve and TPR@FPR metric 
support plot multiple ROC curves with different model
 2020-05-20 14:29:33 +08:00
liaoxingyu e344eae1cc feat: support plotting roc curve and compute auc score 
ROC curve and AUC score will be help for thresholds
 2020-05-19 20:45:26 +08:00
liaoxingyu fd90555e19 feat: add multi-dataset joint training 
new feature that can support joint training, and find some bugs in funtion combine_all of datasets/bases.py that assume person id in dataset has been relabeld from 0 to num_class.
Another bug appears in msmt17 which trainset and testset person id both begin from 0, and we should change testset id from num_class of trainset.
 2020-05-18 20:06:04 +08:00
liaoxingyu 579a5cf552 fix: modify re-rank typro 2020-05-18 17:05:20 +08:00
liaoxingyu d63bf5facc fix: add syncBN options in defaultTraine 2020-05-16 22:44:53 +08:00
liaoxingyu b28c0032e8 fix: add monkey-patching to enable syncBN 
add a trigger to make syncBN work
 2020-05-15 13:33:33 +08:00
liaoxingyu 18a33f7962 feat: add MGN model 
support MGN architecture and training config
 2020-05-15 11:39:54 +08:00
liaoxingyu bf18479541 fix: revise syncBN bug 2020-05-14 14:52:37 +08:00
liaoxingyu 0872a32621 feat: add syncBN support 2020-05-14 13:15:09 +08:00
liaoxingyu 0356ef8c5c feat: add SyncBN and GroupNorm suppor 2020-05-14 11:36:28 +08:00
liaoxingyu 5ae3d4fecf feat: add aqe support in test phase 
query expansion will combine the retrived topk nearest neighbors with the original query feature,
it will enhance mAP by a large margin.feat:
 2020-05-13 16:27:22 +08:00
liaoxingyu 320010f2ae feat: support re-rank in test phase 2020-05-13 11:47:52 +08:00
Xingyu Liao 01d940bfdd
update vehicle dataset 
Summary: Pull Request resolved: #49 

Reviewed By: xingyu liao
 2020-05-12 21:44:05 +08:00
liaoxingyu 9addfb0ae2 feat: support visualizing label list 
add features to support label list visualization, which can be used
for label correction or check the hardest sample
 2020-05-12 21:35:33 +08:00
Jinkai Zheng 640c9bfc97
Add small, medium and large vehicle test datasets 2020-05-11 01:34:19 -05:00
Jinkai Zheng e059b751a6
Add modified vehicle datasets with small, medium and large test datastes 2020-05-11 01:30:56 -05:00
liaoxingyu 9b6fda3830 style: remove title in visualization 2020-05-11 14:12:29 +08:00
liaoxingyu 13bb03eb07 feat: add rank result visualization tools 
Update visualization tools which can save rank list with AP metrics from high to low, vice versa.
In order to compute AP fast in visualizer, modify rank_cylib to get all_AP instead of mAP.
In this way, we can use Cython to compute results.
 2020-05-10 23:17:10 +08:00
liaoxingyu 651e6ba9c4 feat: support multiprocess predictor 
add asyncpredictor to support multiprocessing feature extraction with dataloader
 2020-05-09 18:23:36 +08:00
liaoxingyu 4be4cacb73 fix: add a simple way to reset data prefetcher when resume training 
use data prefetcher build-in reset function to reload it rather than
redefining a new data prefetcher, otherwise it will introduce other
problems in eval-only mode.
 2020-05-09 11:58:27 +08:00
liaoxingyu 9fae467adf feat(engine/defaults): add DefaultPredictor to get image reid features 
Add a new predictor interface, and modify demo code to predict image features.
 2020-05-08 19:24:27 +08:00
liaoxingyu 8ab0bc2455 style(backbone): make parameters loading logging more elegant 2020-05-08 12:22:06 +08:00
liaoxingyu 0b15ac4e03 feat(hooks&optim): update stochastic weight averging hooks 
Update swa method which will do after regular training if you
set this option enabled.
 2020-05-08 12:20:04 +08:00
liaoxingyu afac8aad5d Fix(engine): fix preciseBN dataloader bugs 
preciseBN needs to pass data prefetcher, but now a DataLoader is passed
 2020-05-06 14:26:34 +08:00
liaoxingyu 948af64fd1 feat: add swa algorithm 
Add swa and related config options,
if it is enabled, model will do swa after regular training
 2020-05-06 10:17:44 +08:00
liaoxingyu 6d96529d4c fix(data): fix resume training bug 
fix dataset pid dictionary loading bug when resume training,
data prefetcher will pre-load a batch of data, this will lead to
misalignment of old pid dict and updated pid dict.
We can address this problem by redefine a prefetcher in resume_or_load
 2020-05-05 23:20:42 +08:00
liaoxingyu a2dcd7b4ab feat(layers/norm): add ghost batchnorm 
add a get_norm fucntion to easily change normalization between batchnorm, ghost bn and group bn
 2020-05-01 09:02:46 +08:00