style(tools): update

chore(version): v0.10.0

1. perf(dataset): update CCCF for retrieval task. 2498c99
2. perf(dataloader): update custom_fn. e179099
3. perf(tools): update bash_train.sh/bash_eval.sh use. 2566d78
4. perf(sampler): make train_sampler be shuffle in distributed. 8c1f9f0
5. feat(sampler): new DistributedPKSampler. b789421
6. perf(train.py): add device setting for single GPU training. 1aeb869
7. perf(pk_sampler.py): update. 84d2161
8. perf(triplet_margin_loss): update. 21be529
9. feat(simpleir): add configs/criterion/sampler for triplet training. 426c64c

chore(version): v0.9.3

* New features
  1. perf(retrieval): update knn_rank implementation. 4bf26fb
  2. perf(retrieval): to avoid invalid labels, using sort_idx_list to sort gallery_img_name_list. 484970e
  3. feat(enhancer.py): judge whether to learn PCA parameters on the gallery dataset. f72addb
* Bug fixes
  1. fix(map_for_oxford.py): repair map calculation use self.batch_rank_name_list replace self.query_name_list to participate in the operation. d2b5a59
  2. fix(resnet.py): pretrained: AssertionError: Missing keys when loading pretrained weights: []. 0f6b15e

chore(version): v0.9.2

* New features
  1. build(python): update install requirements. 6d2dea0
  2. perf(evaluate_features.py): add args.query_dir as data_root param in Metric. d29da63
  3. perf(metric): add query_dir as data_root for Metric and return [map] for Oxford. d7f0aa1
  4. feat(metric): add MapForOxford. d00283d
  5. feat(metric): add map. 7ac7fb8
* Bug fixes
  1. fix(mobilenet.py): update load_state_dict_from_url use. 4136efe
  2. fix(dataset): add fake classes and modify gt_root. b2bf843
* Breaking changes

chore(version): v0.9.1

* New features
  1. build(python): update install requires. 110bcbd
  2. perf(train): use torch.cuda.apex and only use cuda:0 to do validate(). 10a87dd
* Bug fixes
  1. fix(sampler): use DistributedSampler only in train mode. bee2f3c
* Breaking changes

chore(version): v0.9.0

* New features
  1. feat(simpleir): add cfg.DATASET.RETRIEVAL_NAME for gallery_dir / query_dir. 41699d8
  2. feat(dataset): add Oxford datset. 0f5a17b
  3. feat(metric): add Precision metric. 2e824d0
  4. feat(configs): add cfg.RETRIEVAL.EXTRACT.PCA_PATH. b55f389
  5. feat(models): add ModelBase. acba468
  6. feat(retrieval): add pca. 7df8201
  7. feat(retrieval): init. a7e977c
  8. feat(distance.py): distance calculation in different ways. 8bd2ba5
* Bug fixes
  1. fix(metric): fix precision(). cb16a53
  2. fix(metric): fix load_retrieval(). 38d1ddb
  3. fix(retrieval): fix rank_name / rank_label saving. a0548f3
  4. fix(simpleir): update. 8552159
  5. fix(extract): import dataset classes. 1f58b69
  6. fix(retrieval): update args usage. b544ed4
  7. fix(extract): update helper.py. d763e92
* Breaking changes
  1. refactor(simpleir): update schema and remove unused code. d799b84
  2. refactor(utils): update. 4dec241

BREAKING CHANGE:

chore(version): v0.8.4

* New features
  1. perf(rerank): add torch qe operation. 41d39dd
  2. refactor(eval): use torch.Tensor replace np.ndarray. 1f0d01a
  3. feat(utils): add torch/numpy argsort use. 3dd48d6
  4. perf(oxford5k): update data split way and save way. 613e17c
  5. feat(eval): add split_file for Caltech101/CUB/Indoor/Oxford5k, derived from pyretri. 9bc8601
  6. feat(split): add Oxford5k support. b38ce79
* Bug fixes
  1. fix(prefetcher): update next() use. 50f8364
  2. fix(prefetcher): make data_prefetcher be Iterator. 707bae1
* Breaking changes

chore(version): v0.8.3

* New features
  1. build(python): upgrade zcls2 ~= 0.4.4 to zcls2 ~= 0.4.5. 00b7ff9
  2. perf(extractor.py): add prefetcher use and update typing hints. 8de4978
* Bug fixes
* Breaking changes

chore(version): v0.8.2

* New features
  1. perf(index): update load_feats() use. 5854d8f
  2. perf(retrieval_images): update retrieval images use. f1f558e
  3. perf(eval.py): add one epoch eval time cal. d3a09ca
  4. perf(eval.py): upgrade pretrained model use. 2ce4c00
  5. perf(index): update max_num and gallery_dict use. a437e9e
* Bug fixes
  1. fix(extract_features.py): fix Extractor import path. 121b466
* Breaking changes

chore(version): v0.8.1

* New features
  1. perf(data): explicitly set val_sampler=None. c46bc96
  2. perf(index): init/clean gallery_dict each epoch and update single mode. e1d8c73
  3. feat(eval.py): new. 2ee957e
* Bug fixes
* Breaking changes