could you provide the google drive link for the resnet weight

Could you please upload the pretrained model in google drive. Thanks you!!!

In training,after 5 epochs I recieved this error "ValueError: too many values to unpack (expected 3)" and I am not able to figure out after a lot of debugging.Screenshot is attached below

After not able to figure out I tried to test your pretained model,this error comes " KeyError: 'nv1.weight' ".Screenshot is attached below.

Can you tell whats wrong in errors
python train.py

python test.py

(url)

hello,have you tried replacing the original ce loss with arcfaceloss? if you have,is the performance better than the one with ce loss？

Can I use your code/model in an open source plugin (MPL-2.0) for a commercial application (Video Management System)?

I tried to visualize the results, but nothing


No feats.pth file appeared at all, and I didn't find it where.
and there is no .cfg.WEIGHT definition in default at all, it is .TEST_WEIGHT? I don’t know what went wrong, help me, there are too many problems.

Hi JackLu,

I really like your code, it's clean yet effective.
I want to adapt se_resnext from official code.
I copy the senet.py file to model/bacbones, then I add the following code in model/make_model.py line 46:

        elif model_name == 'se_resnext50':
            self.in_planes = 2048
            self.base = SENet(block=SEResNeXtBottleneck,
                              layers=[3, 4, 6, 3],
                              groups=32,
                              reduction=16,
                              dropout_p=None,
                              inplanes=64,
                              input_3x3=False,
                              downsample_kernel_size=1,
                              downsample_padding=0,
                              last_stride=last_stride)

I also prepared the ImageNet pre-traind weights and configs.
Then python train.py
The code runs well, but the mAP is not increasing even after many epochs, it stuck in about 0.2%.
I think do not modify other codes, do you have any idea about it?
Thanks a lot!!!

Trying to run the code. Got the error

KeyError: 'base.conv1.weight' in backbone file.

Have any idea to solve this?

Thank you.

已经解决

Hi! Thanks for your great work!
Is harder example mining a method from a published paper? If yes, can you tell me the name of this paper? Thanks

I appreciate your sententious work. However, I failed to reproduce your result under the setting of triplet+softmax, I thought it may result from the hyperparameters ( I only change COS_LAYER->no, LOSS_TYPE->triplet+softmax, TEST_METHOD->euclidean ). So, I hope you can provide the hyperparameters under the setting of triplet+softmax. Thank you!

" raise RuntimeError('Attempting to deserialize object on a CUDA '
RuntimeError: Attempting to deserialize object on a CUDA device but torch.cuda.is_available() is False. If you are running on a CPU-only machine, please use torch.load with map_location='cpu' to map your storages to the CPU."

anyone who can help this? and I change map as below, still report error while run "python train.py".

def load_param(self, trained_path):
    param_dict = torch.load(trained_path,**map_location='cpu'**)

def load_param(self, model_path):
param_dict = torch.load(model_path,map_location='cpu')

Logs folder is missing

can you tell me more about 'flipped feature' ?

你好，谢谢你的实现，问一个比较愚蠢的问题，utils/metrics.py里，函数eval_func里代码matches = (g_pids[indices] == q_pids[:, np.newaxis]).astype(np.int32)是什么意思？matches没一行表示的是什么意思？

还有就是，如果re-id网络在实际运用的时候，是不是只需要计算下待查找的box的特征，然后和检索库中计算所有待检索box的特征相比较，把距离（欧式距离或者余弦距离）最近的前N（1 or 5 or 10)个待检索的box取出来就可以了？

小白一枚，刚接触person-reid，望不吝赐教，感谢

Can you mirror the model to google drive or drop box

File "tools/get_vis_result.py", line 55, in
gallery_feats = torch.load(Cfg.LOG_DIR + 'feats.pth')
File "/home/gyc197/anaconda3/lib/python3.6/site-packages/torch/serialization.py", line 356, in load
f = open(f, 'rb')
FileNotFoundError: [Errno 2] No such file or directory: './logfeats.pth'

hi!
I run your code but reported a strange error:File "./code/triplet_loss.py", line 62, in hard_example_mining dist_ap, relative_p_inds = torch.max(dist_mat[is_pos].contiguous().view(N, -1), 1, keepdim=True)
RuntimeError: shape '[64, -1]' is invalid for input of size 288
I think it for a long time, but I don't know how to fix it, can anyone help me？

do you use the arcfaceloss in the project , but I con't find it any where.

用给出的代码训练出的准确率是85.7%；将loss改为三种相加，也是百分之八十五多；请问有哪些注意事项吗

请问这个怎么解决？谢谢！

File "tools/test.py", line 61, in main
model.load_param(cfg.TEST.WEIGHT)
File "./modeling/baseline.py", line 176, in load_param
for i in param_dict:
TypeError: 'Baseline' object is not iterable

Hello,

I am trying to replicate your code, but apparently my environment is lacking something that I completely ignore.
Can you please provide me with a Dockerfile with all the required dependencies?
Many thanks!

hi, thank you for your good work! It's helpful for me! But I run your code can only get a result of map=86.2% rank-1=94.6%. How to get the best result as you described map=94.2% Rank -1=95.8%？ and in “B1 + Harder Example Mining + searched reranking”， what does "B1" mean?

Nice work!
But I got low results on market dataset using your default config. (I see you get 88% Rank1)

The test feature is normalized
=> Computing DistMat with euclidean distance
2019-06-13 09:21:40,468 baseline-person-reid.train INFO: Validation Results - Epoch: 120
2019-06-13 09:21:40,468 baseline-person-reid.train INFO: mAP: 57.9%
2019-06-13 09:21:40,468 baseline-person-reid.train INFO: CMC curve, Rank-1  :82.7%
2019-06-13 09:21:40,468 baseline-person-reid.train INFO: CMC curve, Rank-5  :93.4%
2019-06-13 09:21:40,468 baseline-person-reid.train INFO: CMC curve, Rank-10 :95.8%

any suggestion ?
I also tested on Duke dataset, but the Rank1 and mAP are still low.

The test feature is normalized
=> Computing DistMat with euclidean distance
2019-06-13 00:16:53,749 baseline-person-reid.train INFO: Validation Results - Epoch: 120
2019-06-13 00:16:53,750 baseline-person-reid.train INFO: mAP: 48.3%
2019-06-13 00:16:53,750 baseline-person-reid.train INFO: CMC curve, Rank-1  :70.3%
2019-06-13 00:16:53,750 baseline-person-reid.train INFO: CMC curve, Rank-5  :83.8%
2019-06-13 00:16:53,750 baseline-person-reid.train INFO: CMC curve, Rank-10 :87.0%

When I training the model using arcface loss,how should I adjust the learning rate?Thank you!