liuwei16 / alfnet Goto Github PK

你好，请问一下论文中2 step的指标是用wma训练吗？为何我利用wma训练得到的结果却并没有达到提供的结果，是有什么其他训练技巧吗？能否回答一下问题，谢谢

I think function of calc_target_multilayer is used to nomal train ,and the calc_target_multilayer_posfirst is used to wma train.am I right?thanks

load weights from data/models/resnet50_weights_tf_dim_ordering_tf_kernels_notop.h5
Epoch 1/1
get_batch_gt: 'NoneType' object has no attribute 'shape'

how to solve this problem ?

If I want to train my data, how should I make a dataset?Need help. I hope to get a reply,thank you.

when I run train.py ,I run into some error
File "/ghome/zhenye/ALFNet-master/keras_alfnet/data_generators.py", line 7, in
from .utils.cython_bbox import bbox_overlaps
ImportError: No module named cython_bbox
besides,
File "/ghome/zhenye/ALFNet-master/keras_alfnet/data_generators.py", line 8, in
from .utils.bbox import box_op
ImportError: No module named bbox
I want to ask the author whether miss the two functions or I miss some important operations?

ImportError: ***cython_bbox.so: undefined symbol: _Py_ZeroStructImportError:
在编译train.py和test.py时都出现了这种问题。请问这是什么原因造成的？是因为cython的版本太新了吗?我的cython版本为0.28 。还是因为cython_bbox.so与py2.7有联系，而我的Python版本是3.7。

when run the train.py ,it will arise Exception . the error informatio are as follow:
File "train_2step_wma.py" line 53 in
model.train_model_wma(C,weight_path,out_path)
File "/workspace/ALFNet/keras_alfnet/model/model_2step.py",line 176 in train_model_wma
print('Exception:{}'. format(e))

anyone could tell me what problems? thank you @liuwei16

Hi, could you provide the training scripts for Caltech dataset? Thank you.

I can't open the MobileNet_v1,who can help me?Thanks

hello, I want to know how to ues this code in video?

I have trained the proposed model on my server with P40 GPU. The MR on Cityperson val dataset is 18% with one GPU (16% for two). Where might the problem arise?

I have trained the proposed model on 2 GTX 1080Ti GPUs (a mini-batch contains 10 images per GPU), which is the setting in the paper. And, I trained for 240k iterations totally, with the initial learning rate of 0.0001 in 160k iteration and degrade the initial learning rate to 0.00001 in the last 80k iterations, but the miss rate is 14.52%. Where might the problem arise?

Can you upload your models' files to Google Drive? I am unable to register to Baidu, as it requires the china number to register

Hi Liu,

Thank you for sharing with the community your excellent work!

I'm trying to reconstruct the mobilenet model, and using the default crop size of 640x1280 px.
I have only 1 GPU with memory of 11.7G (same model as in the paper GTX 1080 Ti).

It seems that this memory size allows me a maximum batch size of 4.
When I try to increase to 10 I get out of memory errors.
How were you able to increase your batch size to 10 ?

Many thanks!!

Hi Liu!

I am highly impressed with your work ! Thanks for publishing the code.

Due to memory issues I've tried to feed the network with smaller data, thus I've defined the crop size in the Config class as self.random_crop = (640, 1152)
instead of (640, 1280). However no it seems that the regression errors are huge!

Do you have an idea why?

Thanks!

follow the code I can train 150 epochs, but when i run the test.py using the train result resnet_e3_l1.15433712553.hdf5 , I cannot get test result, the val_det.txt is empty, why?

@yongqiangzhang1
You can have a try.
nms.zip

hi, @VideoObjectSearch , when i use the nms.zip, i have the problem:
ImportError: libcudart.so.8.0: cannot open shared object file: No such file or directory
i use CUDA9.0, how can i compile to make it work?

I meet the same problem, do you find the cuda 9 version of nms?

you can try nms.zip

I meet the same problem, do you find the cuda 10.1 version of nms?

when I trained the 1 step, can you tell me how to train the 2step, what the lr did you set? And what the epoch did you set?In your paper,the network is totally trained 24k iterations, is that containing 2teps or each step trained 24k iterations?
ok,i have got how to train the model ,but the best results that i get is 12.53 miss rate.

Thank you for your excellent work and great effort to improve performance! I was having a little trouble when I tried to retrain the 2step model using the updated "train_2step_wma.py" file. I set the parameters exactly as described in the paper. In fact, I had not modified any parameters in the program at all and everything remained the default. But I can not get the same results as provided in the description. That makes me very troubled and it must be something important to be ignored. Would you please help me?

The MR I got is 13.25, which corresponds to model “resnet_e80_l0.0647282644082.hdf5”. Reduce learning rate to 1e-5 and 1e-6 can not significantly improve performance. Obviously, there is still a big gap between this and the best results (below 12) provided in “results_2step_wma.txt”.

The other thing I want to confirm is that the download link for the backbone Resnet-50 points to "resnet50_weights_tf_dim_ordering_tf_kernels.h5". In fact, we need "resnet50_weights_tf_dim_ordering_tf_kernels_notop.h5". Does this mean that I should change the file name?

I'm new to detection and find this problem:

from nms.gpu_nms import gpu_nms
ImportError: No module named gpu_nms

it seems like we need to compile the cython code, I don't know how to do it, even though I write a new setup.py but it gets another error...

if I compile it from the faster-rcnn repo, I got the problem that I use gcc-7.3 while they only support the version before 6.

The one released in another issue is only for cuda 8 while I use cuda 9, therefore, I don't know how to fix this problem.

could someone tell me how to fix this problem?

Hello, Mr.liu! I am learning about your paper and test the project you uploaded. However, I got the error that "tensorflow.python.framework.errors_impl.InternalError: cudaGetDevice() failed. Status: CUDA driver version is insufficient for CUDA runtime version". Can you help me with that and sharing the GPU configuration of your successful experiments? Thanks a lot!

I trained the models using four GPU card with default parameters. But I get the results as follows:

Why is this problem? Is there any trick in parameter setting?