visdrone / dronecrowd Goto Github PK

How to divide the training set, validation set, test set？？
And your code don't have make_data.py。。。

Hi,
Why does val_data contain the same images as test_data?

val_data

bartosz@bartosz-pro:~/Downloads/val_data/images$ ll | head -n 20
total 111000
drwxrwxr-x 2 bartosz bartosz  20480 lis  6  2020 ./
drwxrwxr-x 4 bartosz bartosz   4096 lis  6  2020 ../
-rw-rw-r-- 1 bartosz bartosz 296163 mar 13  2018 img011001.jpg
-rw-rw-r-- 1 bartosz bartosz 295048 mar 13  2018 img011002.jpg
-rw-rw-r-- 1 bartosz bartosz 291832 mar 13  2018 img011003.jpg
-rw-rw-r-- 1 bartosz bartosz 292177 mar 13  2018 img011004.jpg
-rw-rw-r-- 1 bartosz bartosz 292308 mar 13  2018 img011005.jpg
-rw-rw-r-- 1 bartosz bartosz 292594 mar 13  2018 img011006.jpg
-rw-rw-r-- 1 bartosz bartosz 291670 mar 13  2018 img011007.jpg
-rw-rw-r-- 1 bartosz bartosz 292245 mar 13  2018 img011008.jpg
-rw-rw-r-- 1 bartosz bartosz 296073 mar 13  2018 img011009.jpg
-rw-rw-r-- 1 bartosz bartosz 293852 mar 13  2018 img011010.jpg
-rw-rw-r-- 1 bartosz bartosz 294563 mar 13  2018 img011011.jpg
-rw-rw-r-- 1 bartosz bartosz 294614 mar 13  2018 img011012.jpg
-rw-rw-r-- 1 bartosz bartosz 294053 lut  9  2019 img015001.jpg
-rw-rw-r-- 1 bartosz bartosz 294591 lut  9  2019 img015002.jpg
-rw-rw-r-- 1 bartosz bartosz 294779 lut  9  2019 img015003.jpg
-rw-rw-r-- 1 bartosz bartosz 294860 mar 13  2018 img015004.jpg
-rw-rw-r-- 1 bartosz bartosz 295132 mar 13  2018 img015005.jpg

test_data

bartosz@bartosz-pro:~/Downloads/test_data/images$ ll | head -n 20
total 2769944
drwxrwxr-x 2 bartosz bartosz 299008 mar  8  2021 ./
drwxrwxr-x 4 bartosz bartosz   4096 mar  8  2021 ../
-rw-rw-r-- 1 bartosz bartosz 296163 mar 13  2018 img011001.jpg
-rw-rw-r-- 1 bartosz bartosz 295048 mar 13  2018 img011002.jpg
-rw-rw-r-- 1 bartosz bartosz 291832 mar 13  2018 img011003.jpg
-rw-rw-r-- 1 bartosz bartosz 292177 mar 13  2018 img011004.jpg
-rw-rw-r-- 1 bartosz bartosz 292308 mar 13  2018 img011005.jpg
-rw-rw-r-- 1 bartosz bartosz 292594 mar 13  2018 img011006.jpg
-rw-rw-r-- 1 bartosz bartosz 291670 mar 13  2018 img011007.jpg
-rw-rw-r-- 1 bartosz bartosz 292245 mar 13  2018 img011008.jpg
-rw-rw-r-- 1 bartosz bartosz 296073 mar 13  2018 img011009.jpg
-rw-rw-r-- 1 bartosz bartosz 293852 mar 13  2018 img011010.jpg
-rw-rw-r-- 1 bartosz bartosz 294563 mar 13  2018 img011011.jpg
-rw-rw-r-- 1 bartosz bartosz 294614 mar 13  2018 img011012.jpg
-rw-rw-r-- 1 bartosz bartosz 294075 mar 13  2018 img011013.jpg
-rw-rw-r-- 1 bartosz bartosz 293981 mar 13  2018 img011014.jpg
-rw-rw-r-- 1 bartosz bartosz 293826 mar 13  2018 img011015.jpg
-rw-rw-r-- 1 bartosz bartosz 294447 mar 13  2018 img011016.jpg
-rw-rw-r-- 1 bartosz bartosz 295887 mar 13  2018 img011017.jpg

In mytest.py it seems to me that calc_trkpt helps you to find the next detection point using tracking. Are you storing tracklets somewhere to help you rebuild the complete track of each pedestrian ? How can we reproduce the tracking that is illustrated in the DroneCrowd README.md ?

Hello everyone!
If someone is looking for a Python version of the "DroneCrowd VID toolkit" for evaluation, check this repository.

在测评每个属性的性能时是怎么划分的？是需要先根据不同选型，选择出来对应数量的图片进行训练和测试，还是直接在所有图片训练完，测试的时候找到对应属性的图片测试？

Hi, the reference to a paper on crowd counting where can I find the implementation codes on training and testing of space-time multi-scale attention network (STANet)?

Hello, thanks for the great work!!

I tried to test the results.
'Setup environment' and 'Download the DroneCrowd data' and 'Ground-Truth Generation' are finished.
The pre-trained models are downloaded.
I get errors when try python mytest.py .
The error ：ValueError: all the input array dimensions for the concatenation axis must match exactly, but along dimension 1, the array at index 0 has size 3 and the array at index 1 has size 2

how might able to solve it?

I am looking forward to your reply.

Hello,
It seems that the spatial-correlation-sampler doesn't work with a RTX 3000 series GPU when we try to install it through pip along with this conda environment
conda create -n STTNet python=3.6 pytorch=1.6 torchvision -c pytorch
I made it work by replacing python and pytorch by newest versions; respectively 3.9 and 1.11.

Do we need oldest versions of pytorch to make STNNet work ?

I have some question about network structure after studying your paper.
(1) In paper, the attention network is used at the localization subnet, however, it is used at the backbone in project.
(2) Why is the output channel of the localized subnet 2? what representative?

How can I calculate the ap@10, ap@15, ap@20 and map in DroneCrowd?

I think the STANet's requirements.txt should change. I think we should add the version because the default version don't support the python2.7. There will be many problem.

https://github.com/VisDrone/DroneVehicle
想问一下这个数据集啥时候能公布，或者先公布一小部分。谢谢

Hi. Do you have a Google Drive link for non-chinese interests? Thank you very much in advance!

感谢作者的杰出贡献，请问一下：DroneCrowd (Full Version)百度云链接什么时候公布？

Thanks for providing such a good data set!
I can't read the files GT_img015203.mat and GT_img015204.mat in test_data/ground_truth both with matlab and python.
I download these files from BaiduYun.
could you update these two files?
thank you so much.