search

Ning-Ding / Implementation-CVPR2015-CNN-for-ReID

Implementation for CVPR 2015 Paper: "An Improved Deep Learning Architecture for Person Re-Identification".
MIT License
147 stars 71 forks source link

The `Merge` layer is deprecated #17

Open zoeyun opened 7 years ago

zoeyun commented 7 years ago

Thank you for share your work. When I trying to run the model using market1501 database, I have problem like this:

File "model_for_market1501.py", line 69, in cross_input_asym tensor_left_padding = K.spatial_2d_padding(tensor_left,padding=(2,2)) File "/usr/local/lib/python2.7/dist-packages/Keras-2.0.3-py2.7.egg/keras/backend/tensorflow_backend.py", line 1834, in spatial_2d_padding assert len(padding[0]) == 2 TypeError: object of type 'int' has no len()

my tensorflow version is 0.12

Thankou for help!!

Ning-Ding commented 7 years ago

Hi, it's my pleasure to share the work. I have updated the model definition script last night, you could use the latest version Keras to define the model now. Today I will complete the model compile part.

Edarke commented 7 years ago

Is there an update on this? I'm still getting "TypeError: object of type 'int' has no len()." Otherwise, is there a last known version of keras/tensorflow in which this worked?

Regardless, I appreciate your work on this implementation.

KeeleyYu commented 6 years ago

Hi, Have you solved this? I met this problem too.

bmiftah commented 6 years ago

The versions of keras and tensorflow worked for my case is :

keras 2.0.4 Tensorflow 1.2.1 Python 2.7.12

I run with CUHK03 But it is incredibly resource intensive .. it takes days with two 11gb GTX1080ti GPU connected with SLI which I enable it going to /etc/X11/xorg.conf and core i9 CPU and RAM size of 32GB ... with min_batch of 150 ( though the author recommended to make it 50 if resource is an issue)

Anybody who complete running the model on CUHK03 ? how long does it take ? what environment/resource you used ?

ChangingFond commented 6 years ago

I completed training the CUHK03 model with two 11G GTX1080ti GPUs, CPU i7-6800K, RAM 32G, tensorflow 1.1, keras 2.0.0. It takes me about 50 hours. @bmiftah