Kaist Dataset , how to get kaist_train_data.npy

[FusionModele]

Can you provide the architecture of kaist_train dataset directory?
and how to generate the kaist_train_data.npy file and what the purpose of this file?
--while trying to run the training file i get this exception----------
2022-04-28 08:29:29.335940: I tensorflow/compiler/xla/service/service.cc:175] StreamExecutor device (0): NVIDIA TITAN RTX, Compute Capability 7.5
load weights from ./data/models/double_resnet.hdf5
Epoch 1/8
try section
get_batch_gt: 'NoneType' object has no attribute 'shape'
get_batch_gt: 'NoneType' object has no attribute 'shape'
get_batch_gt: 'NoneType' object has no attribute 'shape'
get_batch_gt: 'NoneType' object has no attribute 'shape'
get_batch_gt: 'NoneType' object has no attribute 'shape'
get_batch_gt: 'NoneType' object has no attribute 'shape'
get_batch_gt: 'NoneType' object has no attribute 'shape'
get_batch_gt: 'NoneType' object has no attribute 'shape'
get_batch_gt: 'NoneType' object has no attribute 'shape'
get_batch_gt: 'NoneType' object has no attribute 'shape'
Exception: need at least one array to concatenate exception shows
Epoch 2/8
try section
Exception: exception shows
Epoch 3/8
try section
Exception: exception shows
Epoch 4/8
try section
Exception: exception shows
Epoch 5/8
try section
Exception: exception shows
Epoch 6/8
try section
Exception: exception shows
Epoch 7/8
model lr changed to 9.999999747378752e-06
try section
Exception: exception shows
Epoch 8/8
try section
Exception: exception shows
Training complete, exiting.

[CalayZhou]

It seems you do not put the training data in the ./data/kaist，please download it according to ReadMe.md.

train data (download to ./data/kaist)
Baidu Cloud(extract code: ABCD) or KAIST website
test data (download to ./data/kaist_test)
Baidu Cloud(extract code: ABCD) or [Google Drive]

Besides, the grounding truth is saved in kaist_train_data.npy, you can open it with numpy to check the architecture.

[FusionModele]

Hi @CalayZhou, thanks for your reply.
My training data is the following:
data
│ ├── kaist
│ │ ├── annotations
│ │ ├── kaist_train
│ │ ├── splits
is it correct ?
thanks in advance.

[CalayZhou]

Hi~ that's right, 'NoneType' is most likely a problem with the read of training images.

[FusionModele]

How can i fix it please ?
there is something to change in some files or what?

[CalayZhou]

i have not met it before, please make sure all the dependencies are same with the recommendation （Ubuntu18.04, tensorflow1.14, keras2.1, python3.6,cuda10.0,cudnn7.6） and code is unchanged. Besides, check the image read path in kaist_train_data.npy is correct.

[FusionModele]

i have installed the same dependencies.
the extension of images on kaist_train_data.npy is .png while my images are .jpg

[FusionModele]

hi @CalayZhou , i'm trying to test the demo on different image visible and thermal but i don't get a result , he didn't detect any person. the code work only on paired images ,can we test it with unpaired images?
thanks in advance.

[CalayZhou]

hi, the provided model is trained on the KAIST dataset in which images are paired, so the provided model will have severe performance degradation on unpaired images. so you can try：（1）align the unpaired image and test with the provided model. （2）Retraining the model with unpaired images if data is sufficient.

[zhen256]

您好，请问怎么修改kaist_train_data.npy中的图像读取路径？

[CalayZhou]

您好，请问怎么修改kaist_train_data.npy中的图像读取路径？

您好，直接在.npy文件修改相应的路径就可以