Open Ehsan-Yaghoubi opened 4 years ago
michuanhaohao
commented
4 years ago You should write an RAP.py by yourself and confirm that self.num_train_pids == 1295. The training set of Market1501 has 751 IDs, so the self.num_train_pids == 751 in Market1501.
Ehsan-Yaghoubi
commented
4 years ago Thank you for your reply. I have already written a script for RAP dataset.
The same problem is asked in these links but there is no solution for it: https://www.gitmemory.com/issue/KaiyangZhou/deep-person-reid/190/502672175 https://github.com/NVlabs/SPADE/issues/68 https://discuss.pytorch.org/t/invalid-index-at-scatter/30805
Here is the code that I have developed with extra assertions that you suggest. I have brought the logs also. As you can see, the number of features in the last layer is 1295 which is compatible with the number of train IDs. I have resized the images to 128*64 also but it didn't help. I have selected only 751 IDs to see maybe the problem is solved, but it didn't help also.
Any suggestion?
Is the problem with forward function in triplet_loss.py?
I debugged forward function in triplet_loss.py and recieved these shapes:
inputs.shape is : torch.Size([64, 1295])
targets.shape is: torch.Size([64])
log_prob_size od inputs is: torch.Size([64, 1295])
unsqueezed targets.shape is: torch.Size([64, 1])
import re
import os
from .bases import BaseImageDataset
import mat4py
class RAP (BaseImageDataset):
dataset_dir = '/media/ehsan/48BE4782BE476810/AA_GITHUP/Anchor_Level_Paper/RAP_resized_imgs'
RAP_reid_mat_file_path = '/media/ehsan/48BE4782BE476810/AA_GITHUP/Anchor_Level_Paper/rap_annotations/RAP_ReId_information.mat'
def __init__(self, verbose=True, **kwargs):
super(RAP, self).__init__()
self._check_before_run()
train, query, gallery = self._process_dir(self.RAP_reid_mat_file_path, self.dataset_dir)
self.train = train
self.query = query
self.gallery = gallery
if verbose:
print("=> RAP is loaded")
self.print_dataset_statistics(train, query, gallery)
self.num_train_pids, self.num_train_imgs, self.num_train_cams = self.get_imagedata_info(self.train)
self.num_query_pids, self.num_query_imgs, self.num_query_cams = self.get_imagedata_info(self.query)
self.num_gallery_pids, self.num_gallery_imgs, self.num_gallery_cams = self.get_imagedata_info(self.gallery)
assert self.num_train_pids == 1295
def _check_before_run(self):
"""Check if all files are available before going deeper"""
if not os.path.exists(self.dataset_dir):
raise RuntimeError("'{}' is not available".format(self.dataset_dir))
if not os.path.exists(self.RAP_reid_mat_file_path):
raise RuntimeError("'{}' is not available".format(self.RAP_reid_mat_file_path))
def _process_dir(self, mat_dir_path, data_dir):
RAP_reid_information = mat4py.loadmat(filename=mat_dir_path)
RAP_reid_information__ = mat4py.loadmat(filename="/media/ehsan/48BE4782BE476810/AA_GITHUP/Anchor_Level_Paper/rap_annotations/RAP_annotation.mat")
image_names_ = RAP_reid_information["image_name"]
image_names__ = RAP_reid_information__["RAP_annotation"]["name"]
image_names_star = [subsub for sub in image_names__ for subsub in sub]
image_names = [subsub for sub in image_names_ for subsub in sub]
pids = RAP_reid_information["person_id"]
# cams = RAP_reid_information["person_cam"]
# train_ids = RAP_reid_information["train_identity"]
# test_ids = RAP_reid_information["test_identity"]
train_indices_ = RAP_reid_information["train_index"]
train_indices = [subsub for sub in train_indices_ for subsub in sub]
gallery_indices_ = RAP_reid_information["test_index"]
gallery_indices = [subsub for sub in gallery_indices_ for subsub in sub]
query_indices = RAP_reid_information["query_index"]
train = []
ids = []
cam_ids = []
for i1, IMG_index1 in enumerate(train_indices):
person_id = pids[IMG_index1-1][0]
if person_id == -1: continue # junk images are just ignored
img_name = image_names[IMG_index1-1]
img_full_path = os.path.join(data_dir, img_name)
camera = img_name.split("-")[0] # e.g. ['CAM01', '2014', '02', '15', '20140215161032', '20140215162620', 'tarid0', 'frame218', 'line1.png']
camera_id = int(re.findall('\d+',camera)[0])
person_id -= 1 # index should start from 0
camera_id -= 1 # index should start from 0
assert 0 <= person_id <= 2588 # There are 2589 person identities in RAP dataset
assert 0 <= camera_id <= 30 # There are 23 cameras with labels between 1 to 31
cam_ids.append(camera_id)
ids.append(person_id)
train.append((img_full_path, person_id , camera_id))
print(">> Number of train Images: ", len(set(train)))
print(">> Number of train IDs: ",len(set(ids)))
print(">> Number of train Camera IDs: ",len(set(cam_ids)))
query= []
ids = []
cam_ids = []
for i2, IMG_index2 in enumerate(query_indices):
person_id = pids[IMG_index2-1][0]
if person_id == -1: continue # junk images are just ignored
img_name = image_names[IMG_index2-1]
img_full_path = os.path.join(data_dir, img_name)
camera = img_name.split("-")[0] # e.g. ['CAM01', '2014', '02', '15', '20140215161032', '20140215162620', 'tarid0', 'frame218', 'line1.png']
camera_id = int(re.findall('\d+',camera)[0])
person_id -= 1 # index should start from 0
camera_id -= 1 # index should start from 0
assert 0 <= person_id <= 2588 # There are 2589 person identities in RAP dataset
assert 0 <= camera_id <= 30 # There are 23 cameras with labels between 1 to 31
cam_ids.append(camera_id)
ids.append(person_id)
query.append((img_full_path, person_id , camera_id))
print(">>>> Number of query Images: ",len(set(query)))
print(">>>> Number of query IDs: ",len(set(ids)))
print(">>>> Number of query Camera IDs: ",len(set(cam_ids)))
gallery= []
ids = []
cam_ids = []
for i3, IMG_index3 in enumerate(gallery_indices):
person_id = pids[IMG_index3-1][0]
if person_id == -1: continue # junk images are just ignored
img_name = image_names[IMG_index3-1]
img_full_path = os.path.join(data_dir, img_name)
camera = img_name.split("-")[0] # e.g. ['CAM01', '2014', '02', '15', '20140215161032', '20140215162620', 'tarid0', 'frame218', 'line1.png']
camera_id = int(re.findall('\d+',camera)[0])
person_id -= 1 # index should start from 0
camera_id -= 1 # index should start from 0
assert 0 <= person_id <= 2588 # There are 2589 person identities in RAP dataset
assert 0 <= camera_id <= 30 # There are 23 cameras with labels between 1 to 31
cam_ids.append(camera_id)
ids.append(person_id)
gallery.append((img_full_path, person_id , camera_id))
print(">>>>>> Number of gallery Images: ",len(set(gallery)))
print(">>>>>> Number of gallery IDs: ",len(set(ids)))
print(">>>>>> Number of gallery Camera IDs: ",len(set(cam_ids)))
return train, query, gallery
The logs are as follows:
2019-12-05 12:29:22,158 reid_baseline INFO: Using 1 GPUS
2019-12-05 12:29:22,158 reid_baseline INFO: Namespace(config_file='../configs/softmax_triplet.yml', opts=[])
2019-12-05 12:29:22,158 reid_baseline INFO: Loaded configuration file ../configs/softmax_triplet.yml
2019-12-05 12:29:22,158 reid_baseline INFO:
MODEL:
PRETRAIN_CHOICE: 'imagenet'
PRETRAIN_PATH: '/home/eshan/.torch/models/resnet50-19c8e357.pth'
METRIC_LOSS_TYPE: 'triplet'
IF_LABELSMOOTH: 'on'
IF_WITH_CENTER: 'no'
INPUT:
SIZE_TRAIN: [256, 128]
SIZE_TEST: [256, 128]
PROB: 0.5 # random horizontal flip
RE_PROB: 0.5 # random erasing
PADDING: 10
DATASETS:
NAMES: ('RAP')
DATALOADER:
SAMPLER: 'softmax_triplet'
NUM_INSTANCE: 4
NUM_WORKERS: 8
SOLVER:
OPTIMIZER_NAME: 'Adam'
MAX_EPOCHS: 120
BASE_LR: 0.00035
CLUSTER_MARGIN: 0.3
CENTER_LR: 0.5
CENTER_LOSS_WEIGHT: 0.0005
RANGE_K: 2
RANGE_MARGIN: 0.3
RANGE_ALPHA: 0
RANGE_BETA: 1
RANGE_LOSS_WEIGHT: 1
BIAS_LR_FACTOR: 1
WEIGHT_DECAY: 0.0005
WEIGHT_DECAY_BIAS: 0.0005
IMS_PER_BATCH: 64
STEPS: [40, 70]
GAMMA: 0.1
WARMUP_FACTOR: 0.01
WARMUP_ITERS: 10
WARMUP_METHOD: 'linear'
CHECKPOINT_PERIOD: 40
LOG_PERIOD: 20
EVAL_PERIOD: 40
TEST:
IMS_PER_BATCH: 128
RE_RANKING: 'no'
WEIGHT: "path"
NECK_FEAT: 'after'
FEAT_NORM: 'yes'
OUTPUT_DIR: "/media/ehsan/HDD2TB/PersonReIdentification/reid-strong-baseline/reid_on_RAP_RESULTS"
2019-12-05 12:29:22,159 reid_baseline INFO: Running with config:
DATALOADER:
NUM_INSTANCE: 4
NUM_WORKERS: 8
SAMPLER: softmax_triplet
DATASETS:
NAMES: RAP
ROOT_DIR: /media/ehsan/HDD2TB/PersonReIdentification/DATASET_Person_Reidentification/RAP/RAP128x64
INPUT:
PADDING: 10
PIXEL_MEAN: [0.485, 0.456, 0.406]
PIXEL_STD: [0.229, 0.224, 0.225]
PROB: 0.5
RE_PROB: 0.5
SIZE_TEST: [256, 128]
SIZE_TRAIN: [256, 128]
MODEL:
DEVICE: cuda
DEVICE_ID: 0
IF_LABELSMOOTH: on
IF_WITH_CENTER: no
LAST_STRIDE: 1
METRIC_LOSS_TYPE: triplet
NAME: resnet50
NECK: bnneck
PRETRAIN_CHOICE: imagenet
PRETRAIN_PATH: /home/eshan/.torch/models/resnet50-19c8e357.pth
OUTPUT_DIR: /media/ehsan/HDD2TB/PersonReIdentification/reid-strong-baseline/reid_on_RAP_RESULTS
SOLVER:
BASE_LR: 0.00035
BIAS_LR_FACTOR: 1
CENTER_LOSS_WEIGHT: 0.0005
CENTER_LR: 0.5
CHECKPOINT_PERIOD: 40
CLUSTER_MARGIN: 0.3
EVAL_PERIOD: 40
GAMMA: 0.1
IMS_PER_BATCH: 64
LOG_PERIOD: 20
MARGIN: 0.3
MAX_EPOCHS: 120
MOMENTUM: 0.9
OPTIMIZER_NAME: Adam
RANGE_ALPHA: 0
RANGE_BETA: 1
RANGE_K: 2
RANGE_LOSS_WEIGHT: 1
RANGE_MARGIN: 0.3
STEPS: (40, 70)
WARMUP_FACTOR: 0.01
WARMUP_ITERS: 10
WARMUP_METHOD: linear
WEIGHT_DECAY: 0.0005
WEIGHT_DECAY_BIAS: 0.0005
TEST:
FEAT_NORM: yes
IMS_PER_BATCH: 128
NECK_FEAT: after
RE_RANKING: no
WEIGHT: path
>> Number of train Images: 13178
>> Number of train IDs: 1295
>> Number of train Camera IDs: 23
>>>> Number of query Images: 7202
>>>> Number of query IDs: 1294
>>>> Number of query Camera IDs: 23
>>>>>> Number of gallery Images: 13460
>>>>>> Number of gallery IDs: 1294
>>>>>> Number of gallery Camera IDs: 23
=> RAP is loaded
Dataset statistics:
----------------------------------------
subset | # ids | # images | # cameras
----------------------------------------
train | 1295 | 13178 | 23
query | 1294 | 7202 | 23
gallery | 1294 | 13460 | 23
----------------------------------------
Loading pretrained ImageNet model......
Baseline(
(base): ResNet(
(conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
(layer1): Sequential(
(0): Bottleneck(
(conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(layer2): Sequential(
(0): Bottleneck(
(conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(3): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(layer3): Sequential(
(0): Bottleneck(
(conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(3): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(4): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(5): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(layer4): Sequential(
(0): Bottleneck(
(conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
)
(gap): AdaptiveAvgPool2d(output_size=1)
(bottleneck): BatchNorm1d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(classifier): Linear(in_features=2048, out_features=1295, bias=False)
)
Train without center loss, the loss type is triplet
>>>
num_classes: 1295
label smooth on, numclasses: 1295
2019-12-05 12:29:27,672 reid_baseline.train INFO: Start training
/home/eshan/.virtualenvs/baseline_person_reidentification/lib/python3.5/site-packages/torch/optim/lr_scheduler.py:82: UserWarning: Detected call of `lr_scheduler.step()` before `optimizer.step()`. In PyTorch 1.1.0 and later, you should call them in the opposite order: `optimizer.step()` before `lr_scheduler.step()`. Failure to do this will result in PyTorch skipping the first value of the learning rate schedule.See more details at https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate
"https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate", UserWarning)
torch.Size([64, 1295])
torch.Size([64])
Traceback (most recent call last):
File "/media/ehsan/48BE4782BE476810/AA_GITHUP/forked_reid_baseline/tools/train.py", line 156, in <module>
main()
File "/media/ehsan/48BE4782BE476810/AA_GITHUP/forked_reid_baseline/tools/train.py", line 152, in main
train(cfg)
File "/media/ehsan/48BE4782BE476810/AA_GITHUP/forked_reid_baseline/tools/train.py", line 68, in train
start_epoch # add for using self trained model
File "/media/ehsan/48BE4782BE476810/AA_GITHUP/forked_reid_baseline/engine/trainer.py", line 208, in do_train
trainer.run(train_loader, max_epochs=epochs)
File "/home/eshan/.virtualenvs/baseline_person_reidentification/lib/python3.5/site-packages/ignite/engine/engine.py", line 326, in run
self._handle_exception(e)
File "/home/eshan/.virtualenvs/baseline_person_reidentification/lib/python3.5/site-packages/ignite/engine/engine.py", line 291, in _handle_exception
raise e
File "/home/eshan/.virtualenvs/baseline_person_reidentification/lib/python3.5/site-packages/ignite/engine/engine.py", line 313, in run
hours, mins, secs = self._run_once_on_dataset()
File "/home/eshan/.virtualenvs/baseline_person_reidentification/lib/python3.5/site-packages/ignite/engine/engine.py", line 280, in _run_once_on_dataset
self._handle_exception(e)
File "/home/eshan/.virtualenvs/baseline_person_reidentification/lib/python3.5/site-packages/ignite/engine/engine.py", line 291, in _handle_exception
raise e
File "/home/eshan/.virtualenvs/baseline_person_reidentification/lib/python3.5/site-packages/ignite/engine/engine.py", line 272, in _run_once_on_dataset
self.state.output = self._process_function(self, batch)
File "/media/ehsan/48BE4782BE476810/AA_GITHUP/forked_reid_baseline/engine/trainer.py", line 47, in _update
loss = loss_fn(score, feat, target)
File "/media/ehsan/48BE4782BE476810/AA_GITHUP/forked_reid_baseline/layers/__init__.py", line 35, in loss_func
return xent(score, target) + triplet(feat, target)[0]
File "/home/eshan/.virtualenvs/baseline_person_reidentification/lib/python3.5/site-packages/torch/nn/modules/module.py", line 547, in __call__
result = self.forward(*input, **kwargs)
File "/media/ehsan/48BE4782BE476810/AA_GITHUP/forked_reid_baseline/layers/triplet_loss.py", line 155, in forward
targets = torch.zeros(log_probs.size()).scatter_(1, targets.unsqueeze(1).data.cpu(), 1)
RuntimeError: Invalid index in scatter at /pytorch/aten/src/TH/generic/THTensorEvenMoreMath.cpp:564
Process finished with exit code 1
When I reordered the IDs from zero, the error disappeared. I changed the above code as follows:
import re
import os
from .bases import BaseImageDataset
import mat4py
from RAP_script.rap_data_loading import load_reid_data
from RAP_script.rap_data_loading import load_rap_dataset
from config import cfg
class RAP (BaseImageDataset):
dataset_dir = 'RAP_resized_imgs256x256'
rap_mat_file = 'Anchor_level_rap/rap_annotations/RAP_annotation.mat'
rap_json_data = "/media/ehsan/48BE4782BE476810/AA_GITHUP/forked_reid_baseline/RAP_script/rap_data.json"
def __init__(self, root='/media/ehsan/48BE4782BE476810/AA_GITHUP/Anchor_Level_Paper', verbose=True, **kwargs):
super(RAP, self).__init__()
self.dataset_dir = os.path.join(root, self.dataset_dir)
self.rap_mat_file = os.path.join(root, self.rap_mat_file)
self._check_before_run()
train, query, gallery, rap_data = self._process_dir(self.rap_mat_file, self.dataset_dir, relabel= True)
self.train = train
self.query = query
self.gallery = gallery
self.rap_data = rap_data
if verbose:
print("=> RAP is loaded")
self.print_dataset_statistics(train, query, gallery)
self.num_train_pids, self.num_train_imgs, self.num_train_cams = self.get_imagedata_info(self.train)
self.num_query_pids, self.num_query_imgs, self.num_query_cams = self.get_imagedata_info(self.query)
self.num_gallery_pids, self.num_gallery_imgs, self.num_gallery_cams = self.get_imagedata_info(self.gallery)
#assert self.num_train_pids == 1295
def _check_before_run(self):
"""Check if all files are available before going deeper"""
if not os.path.exists(self.dataset_dir):
raise RuntimeError("'{}' is not available".format(self.dataset_dir))
if not os.path.exists(self.rap_mat_file):
raise RuntimeError("'{}' is not available".format(self.rap_mat_file))
def _process_dir(self, rap_mat_file, data_dir, relabel):
image_names, pids, train_indices, gallery_indices, query_indices=load_reid_data(rap_mat_file)
rap_data = load_rap_dataset(rap_attributes_filepath=self.rap_mat_file, rap_keypoints_json=self.rap_json_data, load_from_file=True)
pid_container = set()
for index, IMG_index1 in enumerate(train_indices):
person_id = pids[IMG_index1-1]
if person_id == -1: continue # junk images are just ignored
pid_container.add(person_id)
pid2label = {pid: label for label, pid in enumerate(pid_container)}
train = []
ids = []
cam_ids = []
flag = 0
for index, IMG_index1 in enumerate(train_indices):
try:
person_id = pids[IMG_index1-1]
if person_id == -1: continue # junk images are just ignored
img_name = image_names[IMG_index1-1]
img_labels = rap_data[img_name]['attrs']
img_full_path = os.path.join(data_dir, img_name)
camera = img_name.split("-")[0] # e.g. ['CAM01', '2014', '02', '15', '20140215161032', '20140215162620', 'tarid0', 'frame218', 'line1.png']
camera_id = int(re.findall('\d+',camera)[0])
assert 1 <= person_id <= 2589 # There are 2589 person identities in RAP dataset
assert 1 <= camera_id <= 31 # There are 23 cameras with labels between 1 to 31
cam_ids.append(camera_id)
ids.append(person_id)
if relabel: person_id = pid2label[person_id] # train ids must be relabelled from zero
train.append((img_full_path, person_id, camera_id, img_labels))
except KeyError:
flag += 1
#print(" Warning at Train set {}: information of below image is not found in rap_data.\n{}".format(flag, img_name))
print(" Warning at Train set: information of {} images are not found in rap_data".format(flag))
# print(">> Number of train Images: ", len(set(train)))
# print(">> Number of train IDs: ",len(set(ids)))
# print(">> Number of train Camera IDs: ",len(set(cam_ids)))
query= []
ids = []
cam_ids = []
flag = 0
for i2, IMG_index2 in enumerate(query_indices):
try:
person_id = pids[IMG_index2-1]
if person_id == -1: continue # junk images are just ignored
img_name = image_names[IMG_index2-1]
img_labels = rap_data[img_name]['attrs']
img_full_path = os.path.join(data_dir, img_name)
camera = img_name.split("-")[0] # e.g. ['CAM01', '2014', '02', '15', '20140215161032', '20140215162620', 'tarid0', 'frame218', 'line1.png']
camera_id = int(re.findall('\d+',camera)[0])
assert 1 <= person_id <= 2589 # There are 2589 person identities in RAP dataset
assert 1 <= camera_id <= 31 # There are 23 cameras with labels between 1 to 31
cam_ids.append(camera_id)
ids.append(person_id)
query.append((img_full_path, person_id , camera_id, img_labels))
except KeyError:
flag += 1
#print(" Warning at Query set {}: information of below image is not found in rap_data.\n{}".format(flag, img_name))
print(" Warning at Train set: information of {} images are not found in rap_data".format(flag))
# print(">>>> Number of query Images: ",len(set(query)))
# print(">>>> Number of query IDs: ",len(set(ids)))
# print(">>>> Number of query Camera IDs: ",len(set(cam_ids)))
gallery= []
ids = []
cam_ids = []
flag = 0
for i3, IMG_index3 in enumerate(gallery_indices):
try:
person_id = pids[IMG_index3-1]
if person_id == -1: continue # junk images are just ignored
img_name = image_names[IMG_index3-1]
img_labels = rap_data[img_name]['attrs']
img_full_path = os.path.join(data_dir, img_name)
camera = img_name.split("-")[0] # e.g. ['CAM01', '2014', '02', '15', '20140215161032', '20140215162620', 'tarid0', 'frame218', 'line1.png']
camera_id = int(re.findall('\d+',camera)[0])
assert 1 <= person_id <= 2589 # There are 2589 person identities in RAP dataset
assert 1 <= camera_id <= 31 # There are 23 cameras with labels between 1 to 31
cam_ids.append(camera_id)
ids.append(person_id)
gallery.append((img_full_path, person_id , camera_id, img_labels))
except KeyError:
flag += 1
#print(" Warning at Gallery set {}: information of below image is not found in rap_data.\n{}".format(flag, img_name))
print(" Warning at Train set: information of {} images are not found in rap_data".format(flag))
# print(">>>>>> Number of gallery Images: ",len(set(gallery)))
# print(">>>>>> Number of gallery IDs: ",len(set(ids)))
# print(">>>>>> Number of gallery Camera IDs: ",len(set(cam_ids)))
return train, query, gallery, rap_data
First, I should mention that the code is working for the
Market1501dataset properly.I want to train the network on the RAP dataset. I have created a
rap.pyas to below. I tried to mimic themarket1501.py. When the training starts, it runs to aRunTimeError. I debugged the code, and I think everything goes on exactly likemarket1501.py, but I don't know the reason for this error. The error comes fromforwardfunction intriplet_loss.py, where we have:targets = torch.zeros(log_probs.size()).scatter_(1, targets.unsqueeze(1).data.cpu(), 1)I debugged and realized that the size of
inputsis[8, 1295]and the size oftargetsis[8], (when I debug the code formarket1501, these sizes are[8, 751]and[8], respectively). I should add that,batch sizeis8, number ofidsin RAP and market1501 are1295and751respectively. Changing theimages' sizeandbatch sizedo not help.The code for rap dataset (
rap.py) is as follows:ERROR tracebak:
Logs/initialization:
POSSIBLE PROBLEM: 1) Maybe the problem is that somewhere in the code targets are defined to be
751but thescattersees1295size array. I am a newbie in PyTorch, could you please guide me through it?