heshuting555 / aicity2020_dmt_vehiclereid Goto Github PK

Sorry to bother you. Where can i get the bboxes ?

碰到上面这个问题，提示需要optimizer放在ls_scheduler前面：
“In PyTorch 1.1.0 and later, you should call them in the opposite order: optimizer.step() before lr_scheduler.step().“
请问是调整什么位置呢？

Hi there,

I was trying your pre-trained model(resnet50_ibn_a_80.pth: https://drive.google.com/file/d/14TTv8mEECkRLgtmmTMY9BPc70rhgmXcu/view?usp=sharing) using configuration: configs/baseline_veri_r50_a.yml.

And I made some little modification as I don't want to train it.

import os
from config import cfg
import argparse
from model import make_model


if __name__ == "__main__":
    parser = argparse.ArgumentParser(description="ReID Baseline Training")
    parser.add_argument(
        "--config_file", default="configs/baseline_veri_r50_a.yml", help="path to config file", type=str
    )
    parser.add_argument("opts", help="Modify config options using the command-line", default=None,
                        nargs=argparse.REMAINDER)

    args = parser.parse_args()



    if args.config_file != "":
        cfg.merge_from_file(args.config_file)
    cfg.merge_from_list(args.opts)
    cfg.freeze()

    os.environ['CUDA_VISIBLE_DEVICES'] = cfg.MODEL.DEVICE_ID

    # train_loader, val_loader, num_query, num_classes = make_dataloader(cfg)
    model = make_model(cfg, num_class=80)
    model.load_param('resnet50_ibn_a_80.pth')

    with torch.no_grad():
        print(model(torch.zeros((1, 3, 224, 224))).shape)

However, I cannot load your model. The logging loads like this:

using resnet50_ibn_a as a backbone
Traceback (most recent call last):
  File "try.py", line 30, in <module>
    model = make_model(cfg, num_class=80)
  File "AICITY2020_DMT_VehicleReID/model/make_model.py", line 124, in make_model
    model = Backbone(num_class, cfg)
  File "AICITY2020_DMT_VehicleReID/model/make_model.py", line 66, in __init__
    self.base.load_param(model_path)
  File "AICITY2020_DMT_VehicleReID/model/backbones/resnet_ibn_a.py", line 164, in load_param
    self.state_dict()[i.replace('module.','')].copy_(param_dict[i])
KeyError: 'base.conv1.weight'

Do you have any idea of what is going on? Thanks for your help in advance.

RT, thanks. I just set _C.MODEL.COS_LAYER = True

你好，请问模型设置文件不兼容mmdet 2.x版本怎么解决？

作者你好, 我在阅读test-mining.py的时候,发现这一段难以理解.请问一下,这一段是起到什么作用呢?

    num = 0
    query_index = []
    while num <= 3102:
        if num == 0:
            query_index.append(0)
            num += 1
        max_sum = 0
        index_sum = []
        for index in range(len(distmat)):
            all_sum = 0
            flag = True
            if index not in query_index:
                for index_q in query_index:
                    all_sum += distmat[index][index_q]
                    if distmat[index][index_q] < thresh:
                        flag = False
                if all_sum > max_sum and flag:
                    max_sum = all_sum
                    index_sum.append(index)
        if index_sum == []:
            break
        else:
            index = index_sum.pop()
            query_index.append(index)
            num += 1
            #print(num,'num')

你好！想知道完整的数据集应该在哪里下载呢，另外请问transformer的模型放置在哪里呢

你好，请问一下怎么用训练的模型测试自己的图片images_test

您好，
我在跑您的算法AICITY2020_DMT_VehicleReID的时候
执行test.py文件，报错Traceback (most recent call last):
File "/media/zq/train/code2/AICITY2020/test.py", line 43, in
train_loader, val_loader, num_query, num_classes = make_dataloader(cfg)
File "/media/zq/train/code2/AICITY2020/datasets/make_dataloader.py", line 53, in make_dataloader
dataset = __factory[cfg.DATASETS.NAMES](root=cfg.DATASETS.ROOT_DIR, plus_num_id=cfg.DATASETS.PLUS_NUM_ID)
File "/media/zq/train/code2/AICITY2020/datasets/aic_aicsim.py", line 31, in init
train = self._process_dir(self.train_dir, relabel=True)
File "/media/zq/train/code2/AICITY2020/datasets/aic_aicsim.py", line 71, in _process_dir
_, _, frame2trackID = self._process_track(path=self.train_track_path) #### Revised
File "/media/zq/train/code2/AICITY2020/datasets/aic_aicsim.py", line 121, in _process_track
file = open(path)
FileNotFoundError: [Errno 2] No such file or directory: '/media/zq/train/code2/data/AIC20_track2/AIC20_ReID/train_track_id.txt'
模型是您的预训练模型final_model/0409_3/resnet101_ibn_a_70.pth
cfg文件是 configs/baseline_aic_se.yml
这个train_track_id.txt文件是干啥的呀
为什么为执行测试命令，还需要train的文件

我下载的数据集是2021版本的，是因为数据集2020和2021的版本不一样吗？

Hi everyone,

Sorry for this kind of question, but I saw there is a train_track_id.txt file called at the training stage, but I do not have it, where can I find it?

Thank you in advance.

Hi,
Is there a way in your code to visualize the result of your model on the veri dataset?

python train.py --config_file='configs/baseline_aic.yml' OUTPUT_DIR "('/data/model/0402_5/')" ### here we get the model resnet101_ibn_a_80.pth in the 0402_5. 这个模型似乎训练后没有用到吗？

Hi!
I met an error when I run the code : File "...\triplet_loss.py", line 83, in hard_example_mining dist_an, relative_n_inds = torch.min( IndexError: min(): Expected reduction dim 1 to have non-zero size.
I don't know how to solve it. Can anyone help me?

Hi, I notice that there is no query and gallery ids in the data. In your code, the query vehicle ids are all '2' and cam ids are 1. I do not see how to evaluate models with these labels. Or do I miss something?

Thanks.

你好！

run.sh前面两步：
python train.py --config_file='configs/baseline_aic.yml' OUTPUT_DIR "('/data/model/0402_5/')"
python train.py --config_file='configs/baseline_aic_finetune.yml' OUTPUT_DIR "('/data/model/0402_6/')"

我理解:
第一步是进行pre-train阶段，使用imagenet预训练模型，输出resnet101_ibn_a_80.pth
第二步是finetune阶段，使用resnet101_ibn_a_80.pth进行训练，冻结前两层，输出resnet101_ibn_a_40.pth

但是在baseline_aic_finetune.yml里，配置项
MODEL:
PRETRAIN_CHOICE: 'finetune'
PRETRAIN_PATH: '/data/model/0402_6/resnet101_ibn_a_40.pth'

这个PRETRAIN_PATH是不是写错了，应该是/0402_5/resnet101_ibn_a_80.pth？ 否则在运行这个train之前，这个文件压根没有生成过？

谢谢！

Hi, I want to convert your pre trained model to .onnx and then try in another project. How I can convert .pth to .onnx?

Thanks.

AI City Challenge 2020 Track2 的数据集官网上已经不提供了，而且申请新的数据集也并未回复，请问能提供联系邮件或者开放的数据集吗？

Very nice project ! I compare with https://github.com/michuanhaohao/reid-strong-baseline/blob/master/configs/softmax_triplet.yml , and see some differences such as

IF_LABELSMOOTH: 'off' # origin is 'on'
SIZE_TRAIN: [320, 320] # origin is [256, 128]
OPTIMIZER_NAME: 'SGD' # origin is 'Adam'
MAX_EPOCHS: 40 # origin is 120
BASE_LR: 0.001 # origin is 0.00035
IMS_PER_BATCH: 96 # origin is 64
......

Can you share some experiences about how to choose these hyp? Thanks a lot !!

I tried this project under same dataset and config file, but this project's mAP and R1 is about 1% lower than the origin baseline. Will it be affected by the ignite lib ? Thanks !

I want to test this baseline on the VeRi-776.First I modified the '.yml' to my own path, then I modified 'veri.py',the root was modified to my own path 'root = "/home/zhang/D/LXF/AICITY2020_DMT_VehicleReID/data' in 'init',then I run 'python train.py --config_file='configs/baseline_veri_r50.yml' . the code reports a error "RuntimeError: '../data/VeRi' is not available
'" I ensure my path is right,and the dataset veri is available.why ? thank you

Hi @heshuting555 ！ thanks for your great work!

When I try to train the VeRi-776 dataset, I met a error.

python train.py --config_file='configs/baseline_veri_r50_a.yml' MODEL.DEVICE_ID "('3')" OUTPUT_DIR "('log/')"

the following is the result

Traceback (most recent call last):
  File "train.py", line 62, in <module>
    model = make_model(cfg, num_class=num_classes)
  File "/data/DMT/model/make_model.py", line 126, in make_model
    model = Backbone(num_class, cfg)
  File "/data/DMT/model/make_model.py", line 67, in __init__
    self.base.load_param(model_path)
  File "/data/DMT/model/backbones/resnet_ibn_a.py", line 164, in load_param
    self.state_dict()[i.replace('module.','')].copy_(param_dict[i])
KeyError: 'base.conv1.weight'

I used your weights resnet50_ibn_a_80.pth

Could you please help me? Thank you very much!

Hi, I have a question about loading query and gallery for AI-City dataset.

Why for query and gallery images, you set camid to 1 and pid to 2 in datasets/aic.py?
Is there any reason for that?
Thanks for helping!

def _process_dir_test(self, dir_path, relabel=False, query=True):
        img_paths = sorted(glob.glob(osp.join(dir_path, '*.jpg')))

        dataset = []
        _, _, frame2trackID = self._process_track(path=self.test_track_path)
        for img_path in img_paths:
            camid = 1
            pid = 2
            if query:
                dataset.append((img_path, pid, camid, -1))
            else:
                trackid = frame2trackID[int(img_path[-10:-4])]
                dataset.append((img_path, pid, camid, trackid))
        #print(len(dataset), 'len(dataset)')
        return dataset

您好，请问如何生成track2.txt文件

RT. Thanks!