训练完成后，自己写的demo测试，发现结果不对，计算两幅图像距离后，同一个人和不同人，之间特征距离都比较小 以下是文件脚本的测试结果： 查询图像：/home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0001_c2s1_000301_00.jpg

0 /home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0001_c2s1_000301_00.jpg 两幅图像距离 ： 5.9604645e-08 1 /home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0001_c2s1_000351_00.jpg 两幅图像距离 ： 0.00012886524 2 /home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0001_c2s1_001976_00.jpg 两幅图像距离 ： 0.00013148785 3 /home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0001_c2s1_082596_00.jpg 两幅图像距离 ： 0.00013911724 4 /home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0001_c2s1_109696_00.jpg 两幅图像距离 ： 0.00010842085 5 /home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0001_c2s3_026007_00.jpg 两幅图像距离 ： 0.00014215708 6 /home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0002_c2s1_000301_00.jpg 两幅图像距离 ： 0.00017541647 7 /home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0002_c2s1_000351_00.jpg 两幅图像距离 ： 0.00015747547 8 /home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0002_c2s1_000801_00.jpg 两幅图像距离 ： 0.0002257824 9 /home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0002_c2s1_000976_00.jpg 两幅图像距离 ： 0.00016981363 10 /home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0002_c2s1_068496_00.jpg 两幅图像距离 ： 0.00023156404 11 /home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0002_c2s1_123041_00.jpg 两幅图像距离 ： 0.00023680925 12 /home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0105_c2s1_017426_01.jpg 两幅图像距离 ： 0.0001938343 13 /home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0105_c2s1_017451_04.jpg 两幅图像距离 ： 0.00017023087 14 /home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0105_c2s1_017476_02.jpg 两幅图像距离 ： 0.0001847744 15 /home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0105_c2s1_017601_01.jpg 两幅图像距离 ： 0.00019276142 16 /home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0105_c2s1_025851_02.jpg 两幅图像距离 ： 0.00020557642 17 /home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0105_c2s1_036726_01.jpg 两幅图像距离 ： 0.00016593933 18 /home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0464_c2s1_119341_04.jpg 两幅图像距离 ： 0.000207901 19 /home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0464_c2s1_119466_04.jpg 两幅图像距离 ： 0.00021898746 20 /home/yan/Documents/bg-file/fast-reid-master/tools/deploy/Image/TestImage/C2/0464_c2s1_119466_06.jpg 两幅图像距离 ： 0.00017267466

求大神指教，谢谢

@JinkaiZheng 您的工程文件代码比较全，我想请教您几个问题

1、FREEZE_ITERS: 2000 作用？ 2、max_iter ，是 1个epoch =数据量 / batch 例如一个数据是10个iter , max_iter最大的迭代= epoch数量 * 每个epoch 的迭代

3、DATASETS: NAMES: ("Market1501",) TESTS: ("Market1501",) #要改成DukeMTMC? 这里我用自己的数据做训练格式变成 Market1501 数据格式， 测试数据的评估，我只是把数据读取路径 改成duke的, 测试结果不对，训练收敛 是我改动的方式错了吗？

This guide explains how to train your own custom dataset with fastreid's data loaders.

Before You Start

Following Getting Started to setup the environment and install requirements.txt dependencies.

Train on Custom Dataset

1. Register your dataset (i.e., tell fastreid how to obtain your dataset).

   To let fastreid know how to obtain a dataset named "my_dataset", users need to implement a Class that inherits fastreid.data.datasets.bases.ImageDataset:

   	from fastreid.data.datasets import DATASET_REGISTRY
   	from fastreid.data.datasets.bases import ImageDataset
   
   
   	@DATASET_REGISTRY.register()
   	class MyOwnDataset(ImageDataset):
   		def __init__(self, root='datasets', **kwargs):
   			...
   			super().__init__(train, query, gallery)		
   

   Here, the snippet associates a dataset named "MyOwnDataset" with a class that processes train set, query set and gallery set and then pass to the baseClass. Then add a decorator to this class for registration.

   The class can do arbitrary things and should generate train list: list(str, str, str), query list: list(str, int, int) and gallery list: list(str, int, int) as below.

   	train_list = [
   	(train_path1, pid1, camid1), (train_path2, pid2, camid2), ...]
   
   	query_list = [
   	(query_path1, pid1, camid1), (query_path2, pid2, camid2), ...]
   
   	gallery_list = [
   	(gallery_path1, pid1, camid1), (gallery_path2, pid2, camid2), ...]
   

   You can also pass an empty train_list to generate a "Testset" only with super().__init__([], query, gallery).

   Notice: query and gallery sets could have the same camera views, but for each individual query identity, his/her gallery samples from the same camera are excluded. So if your dataset has no camera annotations, you can set all query identities camera number to 0 and all gallery identities camera number to 1, then you can get the testing results.

2. Import your dataset.

   Aftre registering your own dataset, you need to import it in train_net.py to make it effective.

   	from dataset_file import MyOwnDataset
   

config文件：

`_BASE_: "Base-bagtricks.yml"

MODEL:
  BACKBONE:
    NAME: "build_resnest_backbone"
    DEPTH: "50x"
    WITH_IBN: True
    PRETRAIN: True
    PRETRAIN_PATH: "/home/yantianyi/reid-1.0/pretrained/resnest50.pth"
  HEADS:
    NAME: "EmbeddingHead"
    NORM: "BN"
    WITH_BNNECK: True
    NECK_FEAT: "before"
    POOL_LAYER: "avgpool"
    CLS_LAYER: "linear"
    EMBEDDING_DIM: 512
  LOSSES:
    NAME: ("Cosface",)
    COSFACE:
      MARGIN: 0.25
      GAMMA: 128
      SCALE: 1.0

DATASETS:
  NAMES: ("Alidata",)
  TESTS: ("VeRi",)

INPUT:
  SIZE_TRAIN: [224, 224]
  SIZE_TEST: [224, 224]
  DO_AUTOAUG: False

  CJ:
    ENABLED: True
    PROB: 0.8
    BRIGHTNESS: 0.35
    CONTRAST: 0.35
    SATURATION: 0.35
    HUE: 0.2

DATALOADER:
  PK_SAMPLER: True
  NAIVE_WAY: True
  NUM_INSTANCE: 4
  NUM_WORKERS: 8

SOLVER:
  OPT: "Adam"
  MAX_EPOCH: 60
  BASE_LR: 0.00035
  BIAS_LR_FACTOR: 2.
  WEIGHT_DECAY: 0.0005
  WEIGHT_DECAY_BIAS: 0.0005
  IMS_PER_BATCH: 196
  FP16_ENABLED: True

  SCHED: "WarmupMultiStepLR"
  STEPS: [25, 40]
  GAMMA: 0.1

  WARMUP_FACTOR: 0.01
  WARMUP_ITERS: 10

  CHECKPOINT_PERIOD: 10

OUTPUT_DIR: "/home/yantianyi/logs/resnest50_ali_512_cos"
`

部分log：
`[01/25 17:57:48 fastreid]: Full config saved to /home/yantianyi/logs/resnest50_ali_512_cos/config.yaml
[01/25 17:57:48 fastreid.utils.env]: Using a generated random seed 48316424
[01/25 17:57:48 fastreid.engine.defaults]: Prepare training set
[01/25 17:57:53 fastreid.data.datasets.bases]: => Loaded Alidata in csv format: 
| subset   | # ids   | # images   | # cameras   |
|:---------|:--------|:-----------|:------------|
| train    | 127817  | 1669888    | 1           |
[01/25 17:57:55 fastreid.engine.defaults]: Auto-scaling the num_classes=127817
[01/25 17:57:56 fastreid.modeling.backbones.resnest]: Loading pretrained model from /home/yantianyi/reid-1.0/pretrained/resnest50.pth
[01/25 17:57:56 fastreid.modeling.backbones.resnest]: The checkpoint state_dict contains keys that are not used by the model:
  fc.{weight, bias}
[01/25 17:57:57 fastreid.engine.defaults]: Model:
Baseline(
  (backbone): ResNeSt(
    (conv1): Sequential(
      (0): Conv2d(3, 32, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
      (1): BatchNorm(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (2): ReLU(inplace=True)
      (3): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (4): BatchNorm(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (5): ReLU(inplace=True)
      (6): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
    )
    (bn1): BatchNorm(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (relu): ReLU(inplace=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): BatchNorm(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (conv2): SplAtConv2d(
          (conv): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2, bias=False)
          (bn0): BatchNorm(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (relu): ReLU(inplace=True)
          (fc1): Conv2d(64, 32, kernel_size=(1, 1), stride=(1, 1))
          (bn1): BatchNorm(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (fc2): Conv2d(32, 128, kernel_size=(1, 1), stride=(1, 1))
          (rsoftmax): rSoftMax()
        )
        (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): BatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu): ReLU(inplace=True)
        (downsample): Sequential(
          (0): AvgPool2d(kernel_size=1, stride=1, padding=0)
          (1): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (2): BatchNorm(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): BatchNorm(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (conv2): SplAtConv2d(
          (conv): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2, bias=False)
          (bn0): BatchNorm(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (relu): ReLU(inplace=True)
          (fc1): Conv2d(64, 32, kernel_size=(1, 1), stride=(1, 1))
          (bn1): BatchNorm(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (fc2): Conv2d(32, 128, kernel_size=(1, 1), stride=(1, 1))
          (rsoftmax): rSoftMax()
        )
        (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): BatchNorm(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): BatchNorm(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (conv2): SplAtConv2d(
          (conv): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2, bias=False)
          (bn0): BatchNorm(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (relu): ReLU(inplace=True)
          (fc1): Conv2d(64, 32, kernel_size=(1, 1), stride=(1, 1))
          (bn1): BatchNorm(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (fc2): Conv2d(32, 128, kernel_size=(1, 1), stride=(1, 1))
          (rsoftmax): rSoftMax()
        )
        (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): BatchNorm(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): BatchNorm(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (avd_layer): AvgPool2d(kernel_size=3, stride=2, padding=1)
        (conv2): SplAtConv2d(
          (conv): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2, bias=False)
          (bn0): BatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (relu): ReLU(inplace=True)
          (fc1): Conv2d(128, 64, kernel_size=(1, 1), stride=(1, 1))
          (bn1): BatchNorm(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (fc2): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
          (rsoftmax): rSoftMax()
        )
        (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): BatchNorm(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu): ReLU(inplace=True)
        (downsample): Sequential(
          (0): AvgPool2d(kernel_size=2, stride=2, padding=0)
          (1): Conv2d(256, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (2): BatchNorm(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): BatchNorm(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (conv2): SplAtConv2d(
          (conv): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2, bias=False)
          (bn0): BatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (relu): ReLU(inplace=True)
          (fc1): Conv2d(128, 64, kernel_size=(1, 1), stride=(1, 1))
          (bn1): BatchNorm(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (fc2): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
          (rsoftmax): rSoftMax()
        )
        (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): BatchNorm(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): BatchNorm(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (conv2): SplAtConv2d(
          (conv): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2, bias=False)
          (bn0): BatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (relu): ReLU(inplace=True)
          (fc1): Conv2d(128, 64, kernel_size=(1, 1), stride=(1, 1))
          (bn1): BatchNorm(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (fc2): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
          (rsoftmax): rSoftMax()
        )
        (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): BatchNorm(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): BatchNorm(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (conv2): SplAtConv2d(
          (conv): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2, bias=False)
          (bn0): BatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (relu): ReLU(inplace=True)
          (fc1): Conv2d(128, 64, kernel_size=(1, 1), stride=(1, 1))
          (bn1): BatchNorm(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (fc2): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
          (rsoftmax): rSoftMax()
        )
        (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): BatchNorm(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): BatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (avd_layer): AvgPool2d(kernel_size=3, stride=2, padding=1)
        (conv2): SplAtConv2d(
          (conv): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2, bias=False)
          (bn0): BatchNorm(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (relu): ReLU(inplace=True)
          (fc1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
          (bn1): BatchNorm(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (fc2): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
          (rsoftmax): rSoftMax()
        )
        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): BatchNorm(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu): ReLU(inplace=True)
        (downsample): Sequential(
          (0): AvgPool2d(kernel_size=2, stride=2, padding=0)
          (1): Conv2d(512, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (2): BatchNorm(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): BatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (conv2): SplAtConv2d(
          (conv): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2, bias=False)
          (bn0): BatchNorm(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (relu): ReLU(inplace=True)
          (fc1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
          (bn1): BatchNorm(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (fc2): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
          (rsoftmax): rSoftMax()
        )
        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): BatchNorm(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): BatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (conv2): SplAtConv2d(
          (conv): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2, bias=False)
          (bn0): BatchNorm(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (relu): ReLU(inplace=True)
          (fc1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
          (bn1): BatchNorm(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (fc2): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
          (rsoftmax): rSoftMax()
        )
        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): BatchNorm(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): BatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (conv2): SplAtConv2d(
          (conv): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2, bias=False)
          (bn0): BatchNorm(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (relu): ReLU(inplace=True)
          (fc1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
          (bn1): BatchNorm(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (fc2): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
          (rsoftmax): rSoftMax()
        )
        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): BatchNorm(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): BatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (conv2): SplAtConv2d(
          (conv): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2, bias=False)
          (bn0): BatchNorm(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (relu): ReLU(inplace=True)
          (fc1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
          (bn1): BatchNorm(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (fc2): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
          (rsoftmax): rSoftMax()
        )
        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): BatchNorm(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): BatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (conv2): SplAtConv2d(
          (conv): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2, bias=False)
          (bn0): BatchNorm(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (relu): ReLU(inplace=True)
          (fc1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
          (bn1): BatchNorm(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (fc2): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
          (rsoftmax): rSoftMax()
        )
        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): BatchNorm(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): BatchNorm(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (avd_layer): AvgPool2d(kernel_size=3, stride=1, padding=1)
        (conv2): SplAtConv2d(
          (conv): Conv2d(512, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2, bias=False)
          (bn0): BatchNorm(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (relu): ReLU(inplace=True)
          (fc1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1))
          (bn1): BatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (fc2): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1))
          (rsoftmax): rSoftMax()
        )
        (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): BatchNorm(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu): ReLU(inplace=True)
        (downsample): Sequential(
          (0): AvgPool2d(kernel_size=1, stride=1, padding=0)
          (1): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (2): BatchNorm(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): BatchNorm(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (conv2): SplAtConv2d(
          (conv): Conv2d(512, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), groups=2, bias=False)
          (bn0): BatchNorm(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (relu): ReLU(inplace=True)
          (fc1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1))
          (bn1): BatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (fc2): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1))
          (rsoftmax): rSoftMax()
        )
        (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): BatchNorm(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): BatchNorm(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (conv2): SplAtConv2d(
          (conv): Conv2d(512, 1024, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), groups=2, bias=False)
          (bn0): BatchNorm(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (relu): ReLU(inplace=True)
          (fc1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1))
          (bn1): BatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
          (fc2): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1))
          (rsoftmax): rSoftMax()
        )
        (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn3): BatchNorm(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu): ReLU(inplace=True)
      )
    )
  )
  (heads): EmbeddingHead(
    (pool_layer): AdaptiveAvgPool2d(output_size=1)
    (bottleneck): Sequential(
      (0): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (1): BatchNorm(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    )
    (classifier): Linear(in_features=512, out_features=127817, bias=False)
  )
)

Selected optimization level O1: Insert automatic casts around Pytorch functions and Tensor methods.

Defaults for this optimization level are: enabled : True opt_level : O1 cast_model_type : None patch_torch_functions : True keep_batchnorm_fp32 : None master_weights : None loss_scale : dynamic Processing user overrides (additional kwargs that are not None)... After processing overrides, optimization options are: enabled : True opt_level : O1 cast_model_type : None patch_torch_functions : True keep_batchnorm_fp32 : None master_weights : None loss_scale : dynamic Warning: multi_tensor_applier fused unscale kernel is unavailable, possibly because apex was installed without --cuda_ext --cpp_ext. Using Python fallback. Original ImportError was: ModuleNotFoundError("No module named 'amp_C'",) Warning: apex was installed without --cpp_ext. Falling back to Python flatten and unflatten. ./fastreid/evaluation/rank.py:15: UserWarning: Cython rank evaluation (very fast so highly recommended) is unavailable, now use python evaluation. 'Cython rank evaluation (very fast so highly recommended) is ' ./fastreid/evaluation/roc.py:19: UserWarning: Cython roc evaluation (very fast so highly recommended) is unavailable, now use python evaluation. 'Cython roc evaluation (very fast so highly recommended) is ' Warning: apex was installed without --cpp_ext. Falling back to Python flatten and unflatten. ./fastreid/evaluation/rank.py:15: UserWarning: Cython rank evaluation (very fast so highly recommended) is unavailable, now use python evaluation. 'Cython rank evaluation (very fast so highly recommended) is ' ./fastreid/evaluation/roc.py:19: UserWarning: Cython roc evaluation (very fast so highly recommended) is unavailable, now use python evaluation. 'Cython roc evaluation (very fast so highly recommended) is '`

We keep this issue open to collect feature requests from users and hear your voice. Our monthly release plan is available here.

You can either:

1. Suggest a new feature by leaving a comment.
2. Vote for a feature request with 👍 or be against with 👎. (Remember that developers are busy and cannot respond to all feature requests, so vote for your most favorable one!)
3. Tell us that you would like to help implement one of the features in the list or review the PRs. (This is the greatest thing to hear about!)

V1.4

• [ ] fused LAMB optimizer
• [ ] ZeRO optimizer
• [ ] Gradient accumulation
• [ ] Swin Transformer backbone

V1.3 (May)

• [x] Clip gradient (07b8251ccb9f296e4be5d49fc98efebaa12d679f)
• [x] Vision Transformer backbone (2cabc3428adf5cb3fc0be0f273cc025c4b0a8781)
• [x] prefetch_generator (#456)
• [x] Reduce evaluation memory cost

V1.2 (March)

• [x] Multiple machine training (#425)
• [x] Torch2trt pipeline (#428)
• [x] RepVGG backbone (#429)
• [x] Partial FC(https://github.com/JDAI-CV/fast-reid/commit/44cee30dfc929df2051d1ca56e36dc152866e96b)
• [ ] Visualize activation maps

V1.1 (February)

• [ ] Documents
• [x] ~~RepVGG backbone (#429)~~(delayed to V.12)
• [x] ~~Torch2trt pipeline. (#428)~~(delayed to V1.2)
• [x] NAIC20 winner solution
• [x] Multi-teacher KD
• [x] ~~Partial FC(https://github.com/JDAI-CV/fast-reid/commit/44cee30dfc929df2051d1ca56e36dc152866e96b)~~ (delayed to V1.2)
• [x] reid model with tensorrt network definition APIs

Welcome to join the fast-reid official communication group. If you have any questions, you can communicate with the author or a group of people who are using fast-reid. The number of people in the group is more than 200, please add this wechat to join wechat group.

PS: the author here.

59,60c59
<     pad: 1
<     ceil_mode: false
---
>     pad: 0
2435a2435,2444
>   name: "avgpool1"
>   type: "Pooling"
>   bottom: "relu_blob49"
>   top: "avgpool_blob1"
>   pooling_param {
>     pool: AVE
>     global_pooling: true
>   }
> }
> layer {
2438c2447
<   bottom: "relu_blob49"
---
>   bottom: "avgpool_blob1"
2449c2458
<   top: "batch_norm_blob54"
---
>   top: "output"

https://github.com/JDAI-CV/fast-reid/tree/master/tools/deploy I saw the site above and converted model.

./run_inference.sh What should I do?

When I test my own data, I find that this problem often occurs.

python ./demo/visualize_result.py --config-file './configs/Market1501/AGW_R50.yml' --vis-label --dataset-name 'Market1501' --output 'logs/market1501/agw_R50/agw_market1501_vis' --opts MODEL.WEIGHTS "./logs/market1501/agw_R50/model_final.pth"

assert num_valid_q > 0, 'Error: all query identities do not appear in gallery'
AssertionError: Error: all query identities do not appear in gallery

I don't clearly know why this happens? If you have time, can you tell me how to make your own dataset？

您好，我将我的数据集改成Market1501的命名规则，但是输入 python3 tools/train_net.py --config-file ./configs/Market1501/bagtricks_R50.yml MODEL.DEVICE "cuda:0" 之后，程式码好像就停住无法训练了，请问是什么问题呢? 是因为我的图片数量太少吗? 还是因为我的图片id太少? 或是因为我只有一个camid和一个sequence? 谢谢

@L1aoXingyu 您好，我用的是3月份更新的代码 只用duke 数据训练测试 参数训练 ，configs/DukeMTMC/mgn_R50-ibn.yml 精度没有很高， rank1=82.72 map = 68.37 和您发布的结果差别很大

区别： 预训练这里我直接给的 pretrain_path: home/fastreid/resnet50_ibn_a.pth.tar 后来排查看到 下载的模型 是 v1.0/resnet50_ibn_a-d9d9bb7b.pth

只有这个区别

ys.platform            linux
Python                  3.9.7 (default, Sep 16 2021, 13:09:58) [GCC 7.5.0]
numpy                   1.22.4
fastreid                1.3 @/ssd8/exec/jiaruoran/python/fast-reid-master/./fastreid
FASTREID_ENV_MODULE     <not set>
PyTorch                 1.7.1+cu101 @/ssd7/exec/jiaruoran/anaconda3/lib/python3.9/site-packages/torch
PyTorch debug build     False
GPU available           True
GPU 0,1,2,3             Tesla K80
CUDA_HOME               /ssd1/shared/local/cuda-10.1
Pillow                  8.4.0
torchvision             0.8.2+cu101 @/ssd7/exec/jiaruoran/anaconda3/lib/python3.9/site-packages/torchvision
torchvision arch flags  sm_35, sm_50, sm_60, sm_70, sm_75
cv2                     4.5.5
----------------------  -----------------------------------------------------------------------------------
PyTorch built with:
  - GCC 7.3
  - C++ Version: 201402
  - Intel(R) Math Kernel Library Version 2020.0.0 Product Build 20191122 for Intel(R) 64 architecture applications
  - Intel(R) MKL-DNN v1.6.0 (Git Hash 5ef631a030a6f73131c77892041042805a06064f)
  - OpenMP 201511 (a.k.a. OpenMP 4.5)
  - NNPACK is enabled
  - CPU capability usage: AVX2
  - CUDA Runtime 10.1
  - NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75
  - CuDNN 7.6.3
  - Magma 2.5.2
  - Build settings: BLAS=MKL, BUILD_TYPE=Release, CXX_FLAGS= -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DUSE_VULKAN_WRAPPER -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-sign-compare -Wno-unused-parameter -Wno-unused-variable -Wno-unused-function -Wno-unused-result -Wno-unused-local-typedefs -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-stringop-overflow, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, USE_CUDA=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON, 

data_loader num_worker设为多个时，内存增长的尤其快，num_worker=0时也会持续增长，排除pytorch dataloader问题

• I run a line_profiler on the ranking function and discovered that the line I modified takes 91.4 of the run time of the rank function. By modifying we can save a lot of ranking time.
• For the Jaccard distance, the code is performing a lot of not-needed computation that is increasing the reranking time.

Hi, Xingyu! I integrate the re-ranking into fast-reid. As suggested, the modified code can make users use the GPU-reranking by changing config. I have tested the code in Market1501 with market_sbs_R101-ibn.pth and the result is all right as shown below.

07/31 15:10:57 fastreid.evaluation.reid_evaluation]: Test with gpu real-time rerank setting
[07/31 15:11:05 fastreid.engine.defaults]: Evaluation results for Market1501 in csv format:
[07/31 15:11:05 fastreid.evaluation.testing]: Evaluation results in csv format: 
| Dataset    | Rank-1   | Rank-5   | Rank-10   | mAP   | mINP   | metric   |
|:-----------|:---------|:---------|:----------|:------|:-------|:---------|
| Market1501 | 96.35    | 98.43    | 98.84     | 95.23 | 90.53  | 95.79    |

The method is as follows.

1. Compile the code of gpu re-ranking . cd fastreid/evaluation/extension; sh make.sh
2. When evaluating, we can use GPU-reranking by changing config.

python3 tools/train_net.py --config-file ./configs/Market1501/bagtricks_R50.yml --eval-only \
TEST.GPU_RERANK.ENABLED True MODEL.WEIGHTS /path/to/checkpoint_file MODEL.DEVICE "cuda:0"

The value will be treated as str if there is no type. Then there will be error

Traceback (most recent call last):
  File "demo/visualize_result.py", line 144, in <module>
    query_indices = visualizer.vis_rank_list(args.output, args.vis_label, args.num_vis, args.rank_sort, args.label_sort, args.max_rank)
  File "./fastreid/utils/visualizer.py", line 158, in vis_rank_list
    query_indices = query_indices[:num_vis]
TypeError: slice indices must be integers or None or have an __index__ method

when we add some custom parameter, such as --num-vis 5

Add RGPR data augmentation for person reid(An Effective Data Augmentation for person re-identification(https://arxiv.org/abs/2101.08533)) mgn_repvgg_wo_RGPR(epoch 60): rank1: 90.44, mAP: 81.56 mgn_repvgg_w_RGPR(epoch 60): rank1: 90.66, mAP: 82.16 mgn_repvgg_w_RGPR(epoch 120): rank1: 91.07, mAP: 82.18