joehammer934 / mgn-reid

Person re-identification MGN model based on tensorflow

Mutiple Granularity Network(MGN)-ReId

Reproduction of paper：Learning Discriminative Features with Multiple Granularitiesfor Person Re-Identification

The architecture of MGN

Composition

MGN based on Tensorflow:

• export_pb.py ： export tensorflow pb file；
• export_pb_with_pre _post.py ： export tensorflow pb file containing preprocess and postprocess；
• export_tf_serving_model.py ： export the tensorflow serving model from tensorflow pb；
• export_tf_serving_with_pre_post.py ： export the tensorflow serving model containing preprocess and postprocess from tensorflow pb；

Train

• Download dataset Market1501；
• Parameter initialization using the pytorch model which is trained by seathiefwang(Optional)；
• Set training parameters and training paths in train.py and start training;

Inference

export_pb_with_pre _post.py and export_tf_serving_with_pre_post.py pack these operations into the model.

result

Results without re-ranking on Market-1501

map	rank@1	rank@3	rank@5	rank@10
0.874288	0.947150	0.975653	0.984857	0.990499

Reference

https://github.com/seathiefwang/MGN-pytorch

https://github.com/lwplw/reid-mgn

@ARTICLE{2018arXiv180401438W,
    author = {{Wang}, G. and {Yuan}, Y. and {Chen}, X. and {Li}, J. and {Zhou}, X.},
    title = "{Learning Discriminative Features with Multiple Granularities for Person Re-Identification}",
    journal = {ArXiv e-prints},
    archivePrefix = "arXiv",
    eprint = {1804.01438},
    primaryClass = "cs.CV",
    keywords = {Computer Science - Computer Vision and Pattern Recognition},
    year = 2018,
    month = apr,
    adsurl = {http://adsabs.harvard.edu/abs/2018arXiv180401438W},
    adsnote = {Provided by the SAO/NASA Astrophysics Data System}
}

mgn-reid's Issues

Tensorflow Version

Can you please tell me which tensorflow version you are using in this code?

inference

During evaluation, we both extractthe features corresponding to original images and the horizontallyflipped versions, then use the average of these as the final features.What are the benefits of doing this? @joehammer934

Ａｂｏｕｔ　Ｔｅｎｓｏｒｆｌｏｗ　ａｎｄ　ＣＵＤＡ，　ＧＰＵ

Ｈｉ，　Ｉ　ｒｕｎ　ｙｏｕｒ　ｔｒａｉｎ，ｐｙ　ｃｏｄｅ　ｗｉｔｈ　Ｔｅｎｓｏｒｆｌｏｗ１．１４　ａｎｄ　ＣＵＤＡ１０．０　ｖｅｒｓｉｏｎ，　ａｎｄ　ｇｏｔ　ｔｈｅ　ｉｓｓｕｅ：
`2019-10-17 14:47:22.499384: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1763] Adding visible gpu devices: 0
2019-10-17 14:47:28.415586: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1181] Device interconnect StreamExecutor with strength 1 edge matrix:
2019-10-17 14:47:28.415647: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1187] 0
2019-10-17 14:47:28.415662: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1200] 0: N
2019-10-17 14:47:28.418205: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1326] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 10440 MB memory) -> physical GPU (device: 0, name: TITAN V, pci bus id: 0000:42:00.0, compute capability: 7.0)
2019-10-17 14:47:37.414226: F ./tensorflow/core/kernels/random_op_gpu.h:227] Non-OK-status: CudaLaunchKernel(FillPhiloxRandomKernelLaunch, num_blocks, block_size, 0, d.stream(), gen, data, size, dist) status: Internal: invalid configuration argument

Process finished with exit code 134 (interrupted by signal 6: SIGABRT)`

Ｉ　ａｍ　ｎｏｔ　ｓｕｒｅ　ｉｆ　ｉｔ　ｉｓ　ｔｈｅ　ｃｏｄｅ　ｏｒ　ｔｈｅ　ｔｅｎｓｏｒｆｌｏｗ　ｂｕｇ，　ｃｏｕｌｄ　ｙｏｕ　ｐｌｅａｓｅ　ｇｉｖｅ　ｍｅ　ｓｏｍｅ　ａｄｖｉｃｅｓ　？

池化问题

原始论文在分支层使用的是全局maxpool，博主代码中用的是全局avgpool，想问一下博主是不是实测这样效果更好

Hi, I want to know how to inference one image

value = pt_dict[tf2pt[str(var)]].numpy() KeyError: 'reduction_0.2.weight'

value = pt_dict[tf2pt[str(var)]].numpy()

KeyError: 'reduction_0.2.weight'

can you provide the best ckpt file?

thanks

ValueError: need at least one array to concatenate

KeyError: 'reduction_0.2.weight'

Hi, I run your train.py code with the model named "model_best.pt" to train and got the issue
Traceback (most recent call last):

  File "G:/shenda/project/juesai/MGN-ReId/train.py", line 251, in <module>
    train()
  File "G:/shenda/project/juesai/MGN-ReId/train.py", line 166, in train
    pt_name_path)
  File "G:/shenda/project/juesai/MGN-ReId/train.py", line 97, in restore_model_v2
    value = pt_dict[tf2pt[str(var)]].numpy()
KeyError: 'reduction_0.2.weight'

Maybe I dont't have the pre train model, Could you provide the ptr-train model?
thanks!!

ValueError: need at least one array to concatenate