update models, schedulers and config files

jixunbo · Jan 10, 2021 · a17dd49 · a17dd49
1 parent 8e2d957
commit a17dd49
Show file tree

Hide file tree

Showing 27 changed files with 1,657 additions and 482 deletions.
diff --git a/README.md b/README.md
@@ -12,7 +12,7 @@ List of functions
 - Warm up cosine annealing learning rate
 - Random erasing augmentation
 - Cutout augmentation
-- Batch Drop Block
+- Batch Drop Block and Batch Erasing
 - Label smoothing(Cross Entropy loss)
 - Triplet loss
 - Multi-Simulatity loss
@@ -29,19 +29,20 @@ Inplemented networks:
 - MGN [[link]](https://arxiv.org/abs/1804.01438)
 - Bag of tricks [[link]](http://openaccess.thecvf.com/content_CVPRW_2019/papers/TRMTMCT/Luo_Bag_of_Tricks_and_a_Strong_Baseline_for_Deep_Person_CVPRW_2019_paper.pdf)
 - OSNet [[link]](https://arxiv.org/abs/1905.00953)
-- Batch Drop Block for Person ReID [[link]](https://arxiv.org/abs/1811.07130)
+- Batch Drop Block(BDB) for Person ReID [[link]](https://arxiv.org/abs/1811.07130)
 
 
 ## Getting Started
-The designed architecture is concise and easy explicable, where the file engine.py defines the train/ test process and main.py controls the overall epochs, and the folders model, loss, optimizer including respective parts of neural network.
+The designed code architecture is concise and easy explicable, where the file engine.py defines the train/ test process and main.py controls the overall epochs, and the folders model, loss, optimizer including respective parts of neural network.
 
 The user-friendly command-line module argparse helps us indicate different datasets, networks, loss functions, and tricks as we need, 
 the detailed options/configurations are described in the bottom of this page.
 
 If you don't have any dataset yet, run `git clone https://github.com/jixunbo/ReIDataset.git` to download Market-1501, DukeMTMC, and MOT17.
 
 To inplement Multi-Parts Multi-Channels Network with Multi-Similarity loss, run
-`python [path to repo]/main.py --datadir [path to datasets] --data_train DukeMTMC --data_test DukeMTMC --model MCMP_n --batchid 8 --batchimage 8 --batchtest 32 --test_every 10 --epochs 120 --save '' --decay_type step_50_80_110 --loss 0.5*CrossEntropy+0.5*MSLoss --margin 0.75 --nGPU 1 --lr 3.5e-4 --optimizer ADAM --random_erasing --w_cosine_annealing --if_labelsmooth --feats 512`
+
+`python [path to repo]/main.py --datadir [path to datasets] --data_train dukemtmc --data_test dukemtmc --model MCMP_n --batchid 8 --batchimage 6 --batchtest 32 --test_every 20 --epochs 110 --loss 0.5*CrossEntropy+0.5*MSLoss --margin 0.7 --nGPU 1 --lr 6e-4 --optimizer ADAM --random_erasing --feats 512 --pool avg --save '' --if_labelsmooth --w_cosine_annealing`
 
 Also, using pre-defined config file
 `python [path to repo]/main.py --config [path to repo]/mcmp_config.yaml --save ''`
@@ -52,29 +53,29 @@ Note that, the option '--datadir' is the dataset root, which contains folder Mar
 
 '--data_train' and '--data_test' specify the name of train/test dataset, which we can train on one dataset but test on another dataset.
 
-'--batchid 6' and '--batchimage 8' indicate that each batch contrains 6 persons, each person has 8 different images, totally 48 images.
+'--batchid 8' and '--batchimage 6' indicate that each batch contrains 8 persons, each person has 6 different images, totally 48 images.
 
 '--epochs' is the epochs we'd like to train, while '--test_every 10' means evaluation will be excuted in every 10 epochs, the parameters of network and optimizer are updated after every every evaluation. 
 
-Actually, for the MCMP model we have two kinds of backbone, MCMP_r we use ResNet 50 as backbone, while MCMP_n is OSNet, OSNet contrains much less parameters but could achieve a little bit better performance than ResNet50.
+Actually, for the MCMP model we have two kinds of backbone, MCMP_r we use ResNet 50 or ResNet 50 IBN as backbone, while MCMP_n is OSNet, OSNet contrains much less parameters but could achieve a little bit better performance than ResNet50.
 
 If you would like to re-inplement Bag of Tricks, run
 
-`python [path to repo]/main.py --datadir [path to datasets] --data_train Market1501 --data_test Market1501 --model ResNet50 --batchid 16 --batchimage 4 --batchtest 32 --test_every 10 --epochs 120 --save '' --decay_type step_40_70 --loss 0.5*CrossEntropy+0.5*Triplet --margin 0.3 --nGPU 1 --lr 3.5e-4 --optimizer ADAM --random_erasing --warmup 'constant' --if_labelsmooth`
+`python [path to repo]/main.py --datadir [path to datasets] --data_train Market1501 --data_test Market1501 --model ResNet50 --batchid 16 --batchimage 4 --batchtest 32 --test_every 10 --epochs 120 --save '' --decay_type step_40_70 --loss 0.5*CrossEntropy+0.5*Triplet --margin 0.3 --nGPU 1 --lr 3.5e-4 --optimizer ADAM --random_erasing --warmup 'linear' --if_labelsmooth`
 
 or 
 
 `python [path to repo]/main.py --config [path to repo]/bag_of_tricks_config.yaml --save`
 
-If you would like to re-inplement PCB, run
+If you would like to re-inplement PCB with powerful training tricks, run
 
-`python [path to repo]/main.py --datadir [path to datasets] --data_train Market1501 --data_test Market1501 --model PCB --batchid 8 --batchimage 8 --batchtest 32 --test_every 10 --epochs 120 --save '' --decay_type step_50_80_110 --loss 0.5*CrossEntropy+0.5*MSLoss --margin 0.75 --nGPU 1 --lr 5e-3 --optimizer ADAM --random_erasing --warmup 'constant' --if_labelsmooth --bnneck --parts 3`
+`python [path to repo]/main.py --datadir [path to datasets] --data_train Market1501 --data_test Market1501 --model PCB --batchid 8 --batchimage 8 --batchtest 32 --test_every 10 --epochs 120 --save '' --decay_type step_50_80_110 --loss 0.5*CrossEntropy+0.5*MSLoss --margin 0.7 --nGPU 1 --lr 5e-3 --optimizer ADAM --random_erasing --warmup 'linear' --if_labelsmooth --bnneck --parts 3`
 
 Note that, the option '--parts' is used to set the number of stripes to be devided, original paper set 6.
 
 And also, for MGN model run
 
-`python [path to repo]/main.py --datadir [path to datasets] --data_train Market1501 --data_test Market1501 --model MGN --batchid 16 --batchimage 4 --batchtest 32 --test_every 10 --epochs 120 --save '' --decay_type step_50_80_110 --loss 0.5*CrossEntropy+0.5*Triplet --margin 1.2 --nGPU 1 --lr 2e-4 --optimizer ADAM --random_erasing --warmup 'constant' --if_labelsmooth`
+`python [path to repo]/main.py --datadir [path to datasets] --data_train Market1501 --data_test Market1501 --model MGN --batchid 16 --batchimage 4 --batchtest 32 --test_every 10 --epochs 120 --save '' --decay_type step_50_80_110 --loss 0.5*CrossEntropy+0.5*Triplet --margin 1.2 --nGPU 1 --lr 2e-4 --optimizer ADAM --random_erasing --warmup 'linear' --if_labelsmooth`
 
 If you have pretrained model and config file, run
 
@@ -106,8 +107,8 @@ If you are hard-core player ^ ^ and you'd like to try different models or option
 ### Results
 | Model | Market1501 | DukeMTMC-reID |
 | --- | -- | -- |
-| MCMP_n | 96.3 (90.9) |  91.3 (82.6) |
-| MCMP_r | 95.6 (90.1) |  90.2 (81.5) |
+| MCMP_n | 96.3 (91.3) |  92.0 (83.2) |
+| MCMP_r | 95.8 (90.4) |  90.5 (82.0) |
 | BoT | 94.2 (85.4) |  86.7 (75.8) |
 | PCB | 95.1 (86.3) |  87.6 (76.6) |
 | MGN | 94.7 (87.5) | 88.7 (79.4) |
@@ -155,7 +156,7 @@ Additionally, the evaluation metric method is the same as bag of tricks [repo](h
 
 '--bnneck', action='store_true', if raise, use BNNeck, only for ResNet and PCB.
 
-'--drop_block', action='store_true', if raise, use Batch Drop Block.
+'--drop_block', action='store_true', if raise, use Batch Drop Block, and '--h_ratio 0.3 and --w_ratio 1.0' indicate the erased region on the feature maps. 
 
 '--pool', type=str, default='avg', choose pooling method, options: avg, max.
 
@@ -171,7 +172,9 @@ Additionally, the evaluation metric method is the same as bag of tricks [repo](h
 
 '--gamma', type=float, default=0.1,learning rate decay factor for step decay.
 
-'--warmup', type=str, default='none', learning rate warmup method, options: linear, constant, none.
+'--warmup', type=str, default='constant', learning rate warmup method, options: linear, constant.
+
+'--w_cosine_annealing', action='store_true', if raise, use warm up cosine annealing learning rate scheduler.
 
 '--pcb_different_lr', type=str, default='True', if 'True', use different lr only for PCB, if lr is 5e-3, then lr for classifier is 5e-3, lr for other part is 5e-4.
 

diff --git a/data_v2/datamanager.py b/data_v2/datamanager.py
@@ -8,6 +8,17 @@
 from .transforms import build_transforms
 from .datasets import init_image_dataset, init_video_dataset
 
+"""
+Speed Up Dataloader
+
+"""
+from prefetch_generator import BackgroundGenerator
+
+class DataloaderX(torch.utils.data.DataLoader):
+
+    def __iter__(self):
+        return BackgroundGenerator(super().__iter__())
+
 
 class DataManager(object):
     r"""Base data manager.
@@ -126,7 +137,9 @@ def __init__(self, args):
         targets = args.data_test.lower()
         height = args.height
         width = args.width
-        transforms = ['random_flip', 'random_crop']
+        transforms = ['random_flip', 
+        'random_crop'
+        ]
         norm_mean = [0.485, 0.456, 0.406]
         norm_std = [0.229, 0.224, 0.225]
         use_gpu = not args.cpu
@@ -137,7 +150,7 @@ def __init__(self, args):
         batch_size_test = args.batchtest
         workers = args.nThread
         train_sampler = 'random'
-        cuhk03_labeled = True
+        cuhk03_labeled = False
         cuhk03_classic_split = False
         market1501_500k = False
 
@@ -177,7 +190,8 @@ def __init__(self, args):
             num_instances=num_instances
         )
 
-        self.train_loader = torch.utils.data.DataLoader(
+        # self.train_loader = torch.utils.data.DataLoader(
+        self.train_loader = DataloaderX(
             trainset,
             sampler=train_sampler,
             batch_size=batch_size_train,
@@ -206,7 +220,8 @@ def __init__(self, args):
                 cuhk03_classic_split=cuhk03_classic_split,
                 market1501_500k=market1501_500k
             )
-            self.testloader[name]['query'] = torch.utils.data.DataLoader(
+            # self.testloader[name]['query'] = torch.utils.data.DataLoader(
+            self.testloader[name]['query'] = DataloaderX(
                 queryset,
                 batch_size=batch_size_test,
                 shuffle=False,
@@ -228,7 +243,8 @@ def __init__(self, args):
                 cuhk03_classic_split=cuhk03_classic_split,
                 market1501_500k=market1501_500k
             )
-            self.testloader[name]['gallery'] = torch.utils.data.DataLoader(
+            # self.testloader[name]['gallery'] = torch.utils.data.DataLoader(
+            self.testloader[name]['gallery'] = DataloaderX(
                 galleryset,
                 batch_size=batch_size_test,
                 shuffle=False,

diff --git a/data_v2/datasets/__init__.py b/data_v2/datasets/__init__.py
@@ -11,6 +11,7 @@
     'cuhk03': CUHK03,
     'cuhk03_detected': CUHK03_Detected,
     'cuhk03_labeled': CUHK03_Labeled,
+    'cuhk03_splited': CUHK03_splited,
     'dukemtmc': DukeMTMCreID,
     'msmt17': MSMT17,
     'viper': VIPeR,

diff --git a/data_v2/datasets/image/__init__.py b/data_v2/datasets/image/__init__.py
@@ -14,5 +14,6 @@
 from .cuhk02 import CUHK02
 from .cuhk03_detected import CUHK03_Detected
 from .cuhk03_labeled import CUHK03_Labeled
+from .cuhk03_splited import CUHK03_splited
 from .msmt17 import MSMT17
 from .mot17 import MOT17
diff --git a/data_v2/datasets/image/cuhk03_splited.py b/data_v2/datasets/image/cuhk03_splited.py
@@ -0,0 +1,93 @@
+from __future__ import absolute_import
+from __future__ import print_function
+from __future__ import division
+
+import sys
+import os
+import os.path as osp
+
+from .. import ImageDataset
+from ..utils import mkdir_if_missing, read_json, write_json
+
+
+class CUHK03_splited(ImageDataset):
+    """CUHK03.
+
+    Reference:
+        Li et al. DeepReID: Deep Filter Pairing Neural Network for Person Re-identification. CVPR 2014.
+
+    URL: `<http://www.ee.cuhk.edu.hk/~xgwang/CUHK_identification.html#!>`_
+
+    Dataset statistics:
+        - identities: 1360.
+        - images: 13164.
+        - cameras: 6.
+        - splits: 20 (classic).
+    """
+    dataset_dir = 'CUHK03'
+    #dataset_url = None
+
+    def __init__(self, root='', split_id=0, cuhk03_labeled=False, cuhk03_classic_split=False, **kwargs):
+        self.root = osp.abspath(osp.expanduser(root))
+        self.dataset_dir = osp.join(self.root, self.dataset_dir)
+        #self.download_dataset(self.dataset_dir, self.dataset_url)
+
+        # self.data_dir = osp.join(self.dataset_dir, 'cuhk03_release')
+        # self.raw_mat_path = osp.join(self.data_dir, 'cuhk-03.mat')
+
+        self.imgs_detected_dir = osp.join(self.dataset_dir, 'images_detected')
+        self.imgs_labeled_dir = osp.join(self.dataset_dir, 'images_labeled')
+
+        self.split_classic_det_json_path = osp.join(
+            self.dataset_dir, 'splits_classic_detected.json')
+        self.split_classic_lab_json_path = osp.join(
+            self.dataset_dir, 'splits_classic_labeled.json')
+
+        self.split_new_det_json_path = osp.join(
+            self.dataset_dir, 'splits_new_detected.json')
+        self.split_new_lab_json_path = osp.join(
+            self.dataset_dir, 'splits_new_labeled.json')
+
+        self.split_new_det_mat_path = osp.join(
+            self.dataset_dir, 'cuhk03_new_protocol_config_detected.mat')
+        self.split_new_lab_mat_path = osp.join(
+            self.dataset_dir, 'cuhk03_new_protocol_config_labeled.mat')
+
+        required_files = [
+            self.dataset_dir,
+            # self.data_dir,
+            # self.raw_mat_path,
+            self.split_new_det_mat_path,
+            self.split_new_lab_mat_path
+        ]
+        self.check_before_run(required_files)
+
+        # self.preprocess_split()
+
+        if cuhk03_labeled:
+            split_path = self.split_classic_lab_json_path if cuhk03_classic_split else self.split_new_lab_json_path
+        else:
+            split_path = self.split_classic_det_json_path if cuhk03_classic_split else self.split_new_det_json_path
+
+        splits = read_json(split_path)
+        assert split_id < len(splits), 'Condition split_id ({}) < len(splits) ({}) is false'.format(
+            split_id, len(splits))
+        split = splits[split_id]
+
+        train = split['train']
+        query = split['query']
+        gallery = split['gallery']
+        new_train_list = []
+        new_query_list = []
+        new_gallery_list = []
+        for item in train:
+            new_train_list.append(
+                [osp.join(self.dataset_dir, item[0][31:]), item[1], item[2]])
+        for item in query:
+            new_query_list.append(
+                [osp.join(self.dataset_dir, item[0][31:]), item[1], item[2]])
+        for item in gallery:
+            new_gallery_list.append(
+                [osp.join(self.dataset_dir, item[0][31:]), item[1], item[2]])
+        super(CUHK03_splited, self).__init__(new_train_list,
+                                     new_query_list, new_gallery_list, **kwargs)
diff --git a/data_v2/datasets/image/dukemtmcreid.py b/data_v2/datasets/image/dukemtmcreid.py
@@ -58,7 +58,11 @@ def process_dir(self, dir_path, relabel=False):
         for img_path in img_paths:
             pid, _ = map(int, pattern.search(img_path).groups())
             pid_container.add(pid)
-        pid2label = {pid:label for label, pid in enumerate(pid_container)}
+        ####### modified ########
+        pid2label = {pid:label for label, pid in enumerate(sorted(pid_container))}
+        # pid2label = {pid:label for label, pid in enumerate(pid_container)}
+
+        #########################
 
         data = []
         for img_path in img_paths:

diff --git a/data_v2/datasets/image/market1501.py b/data_v2/datasets/image/market1501.py
@@ -77,7 +77,11 @@ def process_dir(self, dir_path, relabel=False):
             if pid == -1:
                 continue # junk images are just ignored
             pid_container.add(pid)
-        pid2label = {pid:label for label, pid in enumerate(pid_container)}
+        ####### modified #######
+        pid2label = {pid:label for label, pid in enumerate(sorted(pid_container))}
+        # pid2label = {pid:label for label, pid in enumerate(pid_container)}
+
+        ########################
 
         data = []
         for img_path in img_paths:

diff --git a/engine_v3.py b/engine_v3.py
@@ -26,6 +26,7 @@ def __init__(self, args, model, optimizer, scheduler, loss, loader, ckpt):
 
         if torch.cuda.is_available():
             self.ckpt.write_log('[INFO] GPU: ' + torch.cuda.get_device_name(0))
+            # print(torch.backends.cudnn.benchmark)
 
         self.ckpt.write_log(
             '[INFO] Starting from epoch {}'.format(self.scheduler.last_epoch + 1))
@@ -53,7 +54,6 @@ def train(self):
 
             self.optimizer.zero_grad()
             outputs = self.model(inputs)
-
             loss = self.loss.compute(outputs, labels)
 
             loss.backward()
@@ -64,6 +64,8 @@ def train(self):
                 batch + 1, len(self.train_loader),
                 self.loss.display_loss(batch)),
                 end='' if batch + 1 != len(self.train_loader) else '\n')
+            # if batch == 0:
+            #     break
 
         self.scheduler.step()
         self.loss.end_log(len(self.train_loader))
@@ -162,8 +164,6 @@ def extract_feature(self, loader, args):
             inputs, pid, camid = self._parse_data_for_eval(d)
             input_img = inputs.to(self.device)
             outputs = self.model(input_img)
-            # print(outputs.shape)
-            # if args.feat_inference == 'after':
 
             f1 = outputs.data.cpu()
             # flip