svishwa / crowdcount-cascaded-mtl Goto Github PK

Single Image Crowd Counting (CNN-based Cascaded Multi-task Learning of High-level Prior and Density Estimation for Crowd Counting)

License: MIT License

MATLAB 18.49% Python 81.51%

crowdcounting

crowdcount-cascaded-mtl's Introduction

CNN-based Cascaded Multi-task Learning of High-level Prior and Density Estimation for Crowd Counting (Single Image Crowd Counting)

This is implementation of the paper CNN-based Cascaded Multi-task Learning of High-level Prior and Density Estimation for Crowd Counting for single image crowd counting which is accepted at AVSS 2017

Installation

Install pytorch
Clone this repository

git clone https://github.com/svishwa/crowdcount-cascaded-mtl.git

We'll call the directory that you cloned crowdcount-cascaded-mtl ROOT

Data Setup

Download ShanghaiTech Dataset

Dropbox: https://www.dropbox.com/s/fipgjqxl7uj8hd5/ShanghaiTech.zip?dl=0

Baidu Disk: http://pan.baidu.com/s/1nuAYslz
Create Directory

mkdir ROOT/data/original/shanghaitech/

Save "part_A_final" under ROOT/data/original/shanghaitech/
Save "part_B_final" under ROOT/data/original/shanghaitech/
cd ROOT/data_preparation/

run create_gt_test_set_shtech.m in matlab to create ground truth files for test data
cd ROOT/data_preparation/

run create_training_set_shtech.m in matlab to create training and validataion set along with ground truth files

Test

Follow steps 1,2,3,4 and 5 from Data Setup
Download pre-trained model files:

[Shanghai Tech A]

[Shanghai Tech B]

Save the model files under ROOT/final_models
Run test.py

a. Set save_output = True to save output density maps

b. Errors are saved in output directory

Training

Follow steps 1,2,3,4 and 6 from Data Setup
Run train.py

Training with TensorBoard

With the aid of Crayon, we can access the visualisation power of TensorBoard for any deep learning framework.

To use the TensorBoard, install Crayon (https://github.com/torrvision/crayon) and set use_tensorboard = True in ROOT/train.py.

Other notes

During training, the best model is chosen using error on the validation set.
10% of the training set is set aside for validation. The validation set is chosen randomly.
Following are the results on Shanghai Tech A and B dataset:
```
         |     |  MAE  |   MSE  |
         ------------------------
         | A   |  101  |   148  |
         ------------------------
         | B   |   17  |    29  |
```
It may be noted that the results are slightly different from the paper. This is due to a few implementation differences as the earlier implementation was in torch-lua. Contact me if torch models (that were used for the paper) are required.

crowdcount-cascaded-mtl's People

Contributors

Stargazers

Watchers

Forkers

wanquan1992 lavonne1213 huanhuanzhang kixiang pumayhui dimens66 lijian8 greenteahua lareinahe roseoncollar dangwanli zumbalamambo deeplearning2012 jsmilemsj opencvfun willdamon fysho mturkson23 aimeng100 wangyintu giserh hassanksalim youngergao aachenhang ccdump kunlaotou mounirb lcaikk1314 ieee820 jiangyuhe123 baiyancheng20 tikam02 akatsukicc sid231 stefan1123 yxxxxxxxx mozachiu yypeng9468 mawah jereo accasio huojiazhi wujinlonglovezhangmiao1314 pikerbright maohule lengzi ideaplexus egrass chunyu-lin-bjtu james-chou 2571630192 starstylesky alvinzh doche star-driver chankeh ensue1 qing-yuan cravisjan97 lim-0 simon5u wzhings qiyvan xliu117 xiao-xiao-jin zuojianhao cccaklll xichunling wenwenlin sijar-1 dwivy521 ltyong qilicun cyao118 ruiqiren pingzhou3 caozhiwei1994 erichuang qiuqiu-lwf lllovage fasladodo jiankangren zml0926 stevievb masterofcs dimkoss11 xinranzero andyzhou4451 fengzifrank

crowdcount-cascaded-mtl's Issues

count group classification labels

How to make the ‘count group classification labels’ mentioned in the paper？Thank you！

the output image is grayscale

when I run test.py, I found the output image is grayscale while it should be colorful in your paper, what should I do to solve this question.

Hi,what is the label tool do you use?

A question for training

Hello,I ran the code you provided, but I was unable to reproduce the effect of the paper,I can only lower my MAE to 30. I have a question if there are any training technique?

Dataset

Hi,
Could you please share the link to ShanghaiTech dataset?
Thank you in advance.

could you please convert the matlib code to python code?

cuda runtime error!
When trying to run test.py I get the following message:
(pytorch) zhouyao@wb-i36g:~/crowdcount-cascaded-mtl$ python test.py --gpu 0
Pre-loading the data. This may take a while...
Loaded 100 / 182
Completed laoding 182 files
/home/zhouyao/anaconda3/envs/pytorch/lib/python2.7/site-packages/h5py/init.py:36: FutureWarning: Conversion of the second argumen t of issubdtype from float to np.floating is deprecated. In future, it will be treated as np.float64 == np.dtype(float).type.
from ._conv import register_converters as _register_converters
/home/zhouyao/crowdcount-cascaded-mtl/src/crowd_count.py:25: UserWarning: Implicit dimension choice for softmax has been deprecated. Change the call to include dim=X as an argument.
density_cls_prob = F.softmax(density_cls_score)
THCudaCheck FAIL file=/opt/conda/conda-bld/pytorch_1513363039688/work/torch/lib/THC/generic/THCTensorCopy.c line=70 error=77 : an ill egal memory access was encountered
Traceback (most recent call last):
File "test.py", line 45, in
density_map = density_map.data.cpu().numpy()
File "/home/zhouyao/anaconda3/envs/pytorch/lib/python2.7/site-packages/torch/tensor.py", line 35, in cpu
return self.type(getattr(torch, self.class.name))
File "/home/zhouyao/anaconda3/envs/pytorch/lib/python2.7/site-packages/torch/cuda/init.py", line 370, in type
return super(_CudaBase, self).type(*args, **kwargs)
File "/home/zhouyao/anaconda3/envs/pytorch/lib/python2.7/site-packages/torch/_utils.py", line 38, in type
return new_type(self.size()).copy(self, async)
RuntimeError: cuda runtime error (77) : an illegal memory access was encountered at /opt/conda/conda-bld/pytorch_1513363039688/work/t orch/lib/THC/generic/THCTensorCopy.c:70

Any suggestions for what might be wrong.

how to fix the invalid value problem

when i change my own dataset , it comes out with:
D:\project\mtl\src\utils.py:8: RuntimeWarning: invalid value encountered in true_divide
density_map = 255*density_map/np.max(density_map)

dataset download issue

Hi I am getting

for the dropbox link and

for the Baidu link when trying to download the shanghaitech tech ds.
I was wondering if I can download them elsewhere, for example, are these the same as the ones here:
https://www.kaggle.com/datasets/tthien/shanghaitech
which can be used directly or have you guys modified?

UC_FF_50 processing code

Dear svishwa, I am not getting shanghatitech images publicly. I also emailed to the authors but yet I dint get any kind of links of the dataset. I have UCF dataset. Can you please provide me the UC_FF_50 data processing marlab codes, if available?

usage Models

How to use trained model for other crowd dataset images.

How to output the crowd count ?

Hello, when I run test.py I get the grayscale density maps, however I don't find the crowd count like it is outputted in your paper. How do we get it ?

为什么要灰度化

您好！请问为什么得到的结果都是灰度图像，而paper中的是彩色图像，是训练需要吗？

Model download issue

I met a issue when i tried to access to the DropBox. So i would be appreciated that if you can provide the final train model in Baidu disk.

Thank you.

Issues

Hi, I didn't download the Pre-trained model files [Shanghai Tech A], [Shanghai Tech A],Can you seed me the files to the email, My email is [email protected]. Thank you very much

Softmax runtime warning

UserWarning: Implicit dimension choice for softmax has been deprecated. Change the call to include dim=X as an argument.
this is the original code line: density_cls_prob = F.softmax(density_cls_score)
So how do I know whether dim should be taken '0' or '1'

evaluate count do not effective in this network.

epoch: 0, step 525, Time: 0.0011s, gt_cnt: 0.0, et_cnt: 7.1
epoch: 0, step 530, Time: 0.0011s, gt_cnt: 59.5, et_cnt: 6.9
epoch: 0, step 535, Time: 0.0010s, gt_cnt: 0.0, et_cnt: 7.4
epoch: 0, step 540, Time: 0.0010s, gt_cnt: 1.0, et_cnt: 7.2
epoch: 0, step 545, Time: 0.0010s, gt_cnt: 0.0, et_cnt: 6.7
epoch: 0, step 550, Time: 0.0010s, gt_cnt: 2.0, et_cnt: 6.4
epoch: 0, step 555, Time: 0.0010s, gt_cnt: 3.0, et_cnt: 6.0
epoch: 0, step 560, Time: 0.0010s, gt_cnt: 0.0, et_cnt: 6.2
epoch: 0, step 565, Time: 0.0010s, gt_cnt: 3.1, et_cnt: 6.3
epoch: 0, step 570, Time: 0.0010s, gt_cnt: 1.0, et_cnt: 6.0
epoch: 0, step 575, Time: 0.0010s, gt_cnt: 7.6, et_cnt: 5.6
epoch: 0, step 580, Time: 0.0010s, gt_cnt: 0.0, et_cnt: 5.8

It seems like that the evaluate count is continuous,but not appeared in 'mcnn' network. What's more, the density maps draw a blank.

how to test on new images

I have crowd images gathered by a security camera. I have all jpg files. How do I run your code on only jpg files. From what I have seen your code requires '.mat'+'.jpg' files in order to produce output. But in my case I donot have .mat file

Any pointers will really help

Error running test.py

When trying to run test.py I get the following message:
File "test.py", line 4, in
from src.crowd_count import CrowdCounter
File "ROOT/crowdcount-cascaded-mtl/src/crowd_count.py", line 5, in
import network
ModuleNotFoundError: No module named 'network'

Thanks in advance for your help