v1.4.0: added cuhksysu

README.rst

@@ -33,6 +33,7 @@ You can find some research projects that are built on top of Torchreid `here <ht
 
 What's new
 ------------
+- [Apr 2021] ``v1.4.0``: We added the person search dataset, `CUHK-SYSU <http://www.ee.cuhk.edu.hk/~xgwang/PS/dataset.html>`_.  Please see the `documentation <https://kaiyangzhou.github.io/deep-person-reid/>`_ regarding how to download the dataset (it contains cropped person images).
 - [Apr 2021] All models in the model zoo have been moved to google drive. Please raise an issue if any model's performance is inconsistent with the numbers shown in the model zoo page (could be caused by wrong links).
 - [Mar 2021] `OSNet <https://arxiv.org/abs/1910.06827>`_ will appear in the TPAMI journal! Compared with the conference version, which focuses on discriminative feature learning using the omni-scale building block, this journal extension further considers generalizable feature learning by integrating `instance normalization layers <https://arxiv.org/abs/1607.08022>`_ with the OSNet architecture. We hope this journal paper can motivate more future work to taclke the generalization issue in cross-dataset re-ID.
 - [Mar 2021] Generalization across domains (datasets) in person re-ID is crucial in real-world applications, which is closely related to the topic of *domain generalization*. Interested in learning how the field of domain generalization has developed over the last decade? Check our recent survey in this topic at https://arxiv.org/abs/2103.02503, with coverage on the history, datasets, related problems, methodologies, potential directions, and so on (*methods designed for generalizable re-ID are also covered*!).

docs/datasets.rst

@@ -189,6 +189,19 @@ CUHK02 (``cuhk02``)
             P4/
             P5/
 
+CUHKSYSU (``cuhksysu``)
+^^^^^^^^^^^^^^^^^^^^^^^^^^
+- Create a folder named "cuhksysu" under ``$REID``.
+- Download the data to "cuhksysu/" from this `google drive link <https://drive.google.com/file/d/1XmiNVrfK2ZmI0ZZ2HHT80HHbDrnE4l3W/view?usp=sharing>`_.
+- Extract the zip file under "cuhksysu/".
+- The data structure should look like
+
+.. code-block:: none
+    
+    cuhksysu/
+        cropped_images
+
+
 Video Datasets
 --------------
 

torchreid/__init__.py

@@ -2,7 +2,7 @@ from __future__ import print_function, absolute_import
 
 from torchreid import data, optim, utils, engine, losses, models, metrics
 
-__version__ = '1.3.6'
+__version__ = '1.4.0'
 __author__ = 'Kaiyang Zhou'
 __homepage__ = 'https://kaiyangzhou.github.io/'
 __description__ = 'Deep learning person re-identification in PyTorch'

torchreid/data/datasets/__init__.py

@@ -2,7 +2,7 @@ from __future__ import print_function, absolute_import
 
 from .image import (
     GRID, PRID, CUHK01, CUHK02, CUHK03, MSMT17, VIPeR, SenseReID, Market1501,
-    DukeMTMCreID, University1652, iLIDS
+    DukeMTMCreID, University1652, iLIDS, CUHKSYSU
 )
 from .video import PRID2011, Mars, DukeMTMCVidReID, iLIDSVID
 from .dataset import Dataset, ImageDataset, VideoDataset
@@ -19,7 +19,8 @@ __image_datasets = {
     'sensereid': SenseReID,
     'prid': PRID,
     'cuhk02': CUHK02,
-    'university1652': University1652
+    'university1652': University1652,
+    'cuhksysu': CUHKSYSU
 }
 
 __video_datasets = {

torchreid/data/datasets/dataset.py

@@ -35,6 +35,10 @@ class Dataset(object):
     # combineall=True
     _junk_pids = []
 
+    # Some datasets are only used for training, like CUHK-SYSU
+    # In this case, "combineall=True" is not used for them
+    _train_only = False
+
     def __init__(
         self,
         train,
@@ -180,6 +184,9 @@ class Dataset(object):
 
     def combine_all(self):
         """Combines train, query and gallery in a dataset for training."""
+        if self._train_only:
+            return
+
         combined = copy.deepcopy(self.train)
 
         # relabel pids in gallery (query shares the same scope)

torchreid/data/datasets/image/__init__.py

@@ -7,6 +7,7 @@ from .viper import VIPeR
 from .cuhk01 import CUHK01
 from .cuhk02 import CUHK02
 from .cuhk03 import CUHK03
+from .cuhksysu import CUHKSYSU
 from .msmt17 import MSMT17
 from .sensereid import SenseReID
 from .market1501 import Market1501

torchreid/data/datasets/image/cuhksysu.py

@@ -0,0 +1,62 @@
+from __future__ import division, print_function, absolute_import
+import sys
+import os
+import os.path as osp
+import glob
+import copy
+
+from ..dataset import ImageDataset
+
+
+class CUHKSYSU(ImageDataset):
+    """CUHKSYSU.
+
+    This dataset can only be used for model training.
+
+    Reference:
+        Xiao et al. End-to-end deep learning for person search.
+
+    URL: `<http://www.ee.cuhk.edu.hk/~xgwang/PS/dataset.html>`_
+    
+    Dataset statistics:
+        - identities: 11,934
+        - images: 34,574
+    """
+    _train_only = True
+    dataset_dir = 'cuhksysu'
+
+    def __init__(self, root='', **kwargs):
+        self.root = osp.abspath(osp.expanduser(root))
+        self.dataset_dir = osp.join(self.root, self.dataset_dir)
+        self.data_dir = osp.join(self.dataset_dir, 'cropped_images')
+        
+        # image name format: p11422_s16929_1.jpg
+        train = self.process_dir(self.data_dir )
+        query = [copy.deepcopy(train[0])]
+        gallery = [copy.deepcopy(train[0])]
+
+        super(CUHKSYSU, self).__init__(train, query, gallery, **kwargs)
+
+    def process_dir(self, dirname):
+        img_paths = glob.glob(osp.join(dirname, '*.jpg'))
+        num_imgs = len(img_paths)
+
+        # get all identities:
+        pid_container = set()
+        for img_path in img_paths:
+            img_name = osp.basename(img_path)
+            pid = img_name.split('_')[0]
+            pid_container.add(pid)
+        pid2label = {pid: label for label, pid in enumerate(pid_container)}
+
+        num_pids = len(pid_container)
+
+        # extract data
+        data = []
+        for img_path in img_paths:
+            img_name = osp.basename(img_path)
+            pid = img_name.split('_')[0]
+            label = pid2label[pid]
+            data.append((img_path, label, 0)) # dummy camera id
+
+        return data