/ML/Crowd-count/crowdcount-mcnn$ python3 test.py
Pre-loading the data. This may take a while...
Traceback (most recent call last):
File "test.py", line 41, in
data_loader = ImageDataLoader(data_path, gt_path, shuffle=False, gt_downsample=True, pre_load=True)
File "/home/tikam/ML/Crowd-count/crowdcount-mcnn/src/data_loader.py", line 35, in init
img = cv2.resize(img,(wd_1,ht_1))
TypeError: integer argument expected, got float

Do you know what the annotation information of the shanghaitech dataset refers to?
I want to know what kind of software can be used for such labeling?

what these two codes mean in line 34 and 35?
ht_1 = int((ht/4)*4)
wd_1 = int((wd/4)*4)

Dear svishwa, I don't know how to devided train and test dataset of UCF_CC_50, could you please provide me the UCF_CC_50 data prepare codes, if available?

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?

This document './data/original/shanghaitech/part_B_final/test_data/ground_truth_csv/' is used in test.py.
But the dataset I got doesn't contain this document.
Can anyone tell me how can I get this document?
Thanks

I'm confused about what wrong with the code, I just use your code to train my model, and select the model which has lowest MAE and MSE as test model, but when I test my mode, I gets higher MAE(195.83) and MSE(265.86) than you provide

density_map = net(im_data, gt_data) # in test.py, this is the density map
the sum of the density map is the count

Originally posted by @winnielbx in #32 (comment)

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?

Could you tell me what is the meaning of 'annPoints = image_info{1}.location'？
I used the command ‘help image_info’ in matlab, but didn't find the answer to the code.

When trying to run test.py I get the following message:
File "C:\Users\GCZX\Desktop\crowdcount-mcnn-master\src\data_loader.py", line 35, in
init
img = cv2.resize(img,(wd_1,ht_1))

TypeError: integer argument expected, got float

good evening,
i don't seem to understand the structure and the content of the .mat files. in the dataset.
i believe it contains ground truth (counting) for images. i know at some point in CNN we need to calculate the error between the network result and the ground truth we have
i try to print one file:
{'image_info': array([[array([[(array([[ 855.32345978, 590.49587357], [ 965.5908524 , 472.79472415], [ 937.09478464, 400.93507502], ..., [ 42.5852337 , 359.87860699], [1017.48233659, 8.99748811], [1017.48233659, 23.31916643]]), array([[920]], dtype=uint16))]], dtype=[('location', 'O'), ('number', 'O')])]], dtype=object), '__version__': '1.0', '__header__': 'MATLAB 5.0 MAT-file, Platform: PCWIN64, Created on: Fri Nov 18 20:06:05 2016', '__globals__': []}

can someone explain whats in these files and how or for what that content is used in crowd estimation.

Can I train again based on the last model or pre-model that I saved?

Is the batch_size=1 when training?

Hi svishwa,
Is there any script take an input image and return estimation map or count number?

Is this model accurate on sparse image?

` if( w <= 2wn2 )
im = imresize(im,[ h,2wn2+1]);
annPoints(:,1) = annPoints(:,1)2wn2/w;
end
if( h <= 2hn2)
im = imresize(im,[2hn2+1,w]);
annPoints(:,2) = annPoints(:,2)2hn2/h;
end```

Can you tell me the function about the episode?
According to the complete code,I think it is unuse for running

hello,I want to train own dataset,but I donot make gt, Looking forward to help,Thanks

I am chinese ,i can't open the link of [Shanghai Tech A] and [Shanghai Tech B]. maybe something woring with the network. is there any help you can do. here is my email ; [email protected]. would u please send me the pre-trained model files. I will be very grateful

Hi svishwa,
I wonder why there're no relu layer in your model? (authors of "Single Image Crowd Counting via Multi Column Convolutional Neural Network" said that relu is adopted in their network.)

hello, I saved the gt_density data as .jpg file ,then convert it to lmdb, and then train in caffe , but the loss is huge, like as
Iteration 9000 (5.64897 iter/s, 177.024s/1000 iters), loss = 3.87866e+06

how can i solve the problem, or something i did is wrong?

Dear Concern,

It will be a great help if you please provide me the UCF_CC_50 pre-processing matlab code for generating test and train data. And UCSD pre-processing matlab code. I tried to modified your shanghatech code, but I think I am doing some mess. Because I have 50 data in the ucf dataset. and 40 is using to train and rest are using to test. Now, those 40 are converted into 360 input images which is small in number and only 10 test data. Please provide me the matlab code of both dataset.

Thank you very much for your active responses.
Asif : [email protected]

lines 49 -61:

`

wn2 = w/8; hn2 = h/8;
wn2 =8 * floor(wn2/8);
hn2 =8 * floor(hn2/8);

annPoints =  image_info{1}.location;
if( w <= 2*wn2 )
    im = imresize(im,[ h,2*wn2+1]);
    annPoints(:,1) = annPoints(:,1)*2*wn2/w;
end
if( h <= 2*hn2)
    im = imresize(im,[2*hn2+1,w]);
    annPoints(:,2) = annPoints(:,2)*2*hn2/h;
end

`
why the if condition w <= 2wn2 or h <= 2hn2 is used? since line 49-51 says w >= 8*wn2. I am quite confused here.

Hi @svishwa,

Can you please suggest me a method to perform crowd counting Real Time from CCTV Camera? using Single Image Crowd Counting via MCNN.

Can I read the CCTV frames and Write it as a Single Image(only one jpg file) continuously, with a time diff and make the crowd count algorithm to count the people in the image??

Thanks
Guru

den = cv2.resize(ht_1,wd_1)
den = den*((htwd)/(ht_1wd_1))
And i couldn't understand what is this meaning?

Hello author, why is the value of epoch not found in the train file?

in file "get_density_map_gaussian.m"
it seem that you have a fix value of sigma to 4.0 (line 19)
whereas in the paper the sigma are change according to mean KNN distance for each head? (session 2.2)
am I missing something of it really difference?

I am getting MAE MSE much greater than 1.07 and 1.35 respectively (given in Zhang et al's paper) for UCSD dataset with your code. Please suggest me what the problem can be. What results did you get with the UCSD dataset?

If there is someone run it successfully by python 3 and tell the experience, I think it will help green people very much.

my operating system is windows 10. It looks like I can't install pytorch in python 2.7 which means I have to change the .py file from python 2.x into python 3.x, but there are many errors.

Thanks for any help!

HI,
Dose anyone know there is a python version for script get_density_map_gaussian?
Thans

Dear,
You previously helped me with sending the Shanghaitech dataset. Now, I am need of Wroldexpo'10 dataset and UCSD dataset.

Let me explain you a scenario about UCF_CC_50 dataset. I have 50 images. Now I put first 40 images in to train folder like shanghaitech, then rest 10 images i put on test folder. finally i got 359 formatted images to train my model. I want to know how authors tested their method with UCF_CC_50 dataset.

Can you please reply me with these 3 queries. Thanks in advance.
[email protected] is my gmail id

Traceback (most recent call last):
File "train.py", line 106, in
density_map = net(im_data, gt_data)
File "/usr/local/lib/python3.7/site-packages/torch/nn/modules/module.py", line 489, in call
result = self.forward(*input, **kwargs)
File "/Users/yinweicheng/Desktop/job/summer-project/crowdcount-mcnn/src/crowd_count.py", line 17, in forward
im_data = network.np_to_variable(im_data, is_cuda=False, is_training=self.training)
AttributeError: module 'network' has no attribute 'np_to_variable'

Hi Vishwanath, look forward to playing around with your code a bit but noticed there's no license listed. Would you mind adding one, ideally MIT or Apache 2.0? Thanks!

i can't open the link of [Shanghai Tech A] and [Shanghai Tech B],here is my emil:[email protected],would u please send me the pre-trained model files .Thanks anyway.

As I can see there is only conv layers and no fully connected (sure) so we could feed image of any size. I have no dataset you trained the model so my question is in topic ;) What sizes of image were used to train the model?

Hello,
I was wondering if the ShahghaiTech crowd counting is publicly available ?

Thanks

Why is the density map I tested is grayscale? Is there any way to get the thermal map?

I rewrite it in caffe,but I can not get that good results,in your code ,when I use epoch 0 model to test,MSE can reach 445,while mine in caffe epoch 0 model is 11535.Have you use pretrain??

Is there any sort normalization of the input done during the training i.e. mean subtraction or centralization of the data. I see that you have implemented batch normalization layer in the network which helps to get rid of normalization. However, I also saw that you have kept it off during training.

Also, how was the loss profile when you trained the network. Was it very less i.e. in the orders of 10^-3 and so on?

Hi can you share your shanghai data set?
I downloaded this data set from http://cs-chan.com/downloads_crowd_dataset.html but got a totally different format with your preprocess.

Hi, I am a new student of deep learning, now we are going to do a headcount function, how can I get the headcount of each picture? Thank you very much for your reply.

Hello author, I want to make a data set myself. I now take 200 images from a video, use the labelimg to get the .xml file, and then use the matlab program to extract the human coordinates to get the .mat file. When I use your program to convert the density map, the partial density map I get is empty, what is the problem. And I want to ask if the spread parameter σ mentioned in the paper are not obtained by k nearest neighbors, why is it set to 4 in your program?

What's the sigma in ShanghaiTech PartB?

hi @svishwa , in the paper,the author mentioned that he down-sampled each training image by 1/4 before generating its density map(in the Remarks),but i did`t see this part in your code?

Besides,what is these code mean in the data_loader.py?

den = cv2.resize(den,(wd_1,ht_1))                
den = den * ((wd*ht)/(wd_1*ht_1))

looking forward to your reply,thanks!!

Hi,what is your label tool?

Why UPG is used but the speed is not improved?