cli98 / dmnet Goto Github PK

When I use fast_rcnn, I don't know the config file and encounter the error message, and I hope I can get your help.

hello
thank you for implementation sharing
I want to test model on video or image. did you share your model? if yes, which file is for testing and inference?
thank you.

Could you please share the train/test split of the UAVDT dataset? I didn't find it on its official website and I am trying to conducting experiments for object detection on this dataset. Thanks a lot !!!

Hi,
I have read the base paper but cannot understand how you applied mcnn for generating density maps. It will greatly help if you provide the mcnn_train file which you have used for training.

thanks writer , this work is very greatful ,but when i want to try study, I found that I can't download the pretrain weight of MCNN ,could writer put the file again by baidu disk？

Hi @Cli98,
I tried to run your code on VisDrone dataset. Could you show me the height and width of the sliding window? And the threshold is set to be 0.08?

can you share the code about uavdt format to json or xml format transform code?

Hi @Cli98,

I tried to use your pretrained weights to inference density map. However, the density map generated by MCNN is scaled down by 4 so I can't do image cropping with this density map (shape mismatch).

Could you please show me how to fix this error?!

Thank you for your excellent work. I have trouble with generating the estimated density map in both 'val' and 'test-dev' subset of VisDrone. All of the values in the estimated density maps of MCNN are below 0.08. In other words, no object-containing regions are activated. May you share your script for estimating density maps using MCNN?

And also, have you evaluated the performance on 'test-dev' subset? It seems that you only evaluated the 'val' subset in your paper, but the label of 'test-dev' subset is also available.

Thanks a lot.

User
你好 请问MCNN的标注文件必须使用.mat吗 其他形式可以吗 比如.txt DMnet 使用的.txt形式的数据集格式吗

Excuse me.
I glance the whole code fastly but i don't see the code for training this model.
If i want to train this model by myself , how can i do this .Thanks

In the Generate_densitty_map_official.py file, why does the coord array in the format_label function have 6 elements? And why do you mention 4coord+1class in the code.

 `if len(coord) != 6:`
       `# 4 coord + 1 class`
       `print("Failed to parse annotation!")`
       `exit()`

Looking forward to your reply.

How to set parameters in Generate_density_map_official.py

train the detection network , you use validation set or train set?

我想问一下，第二步已经裁剪过图片，并且保存过标注信息了，为什么第三步进行目标检测create_VOC_annotation_official.py转成voc格式时还有裁剪图片的操作，不是直接对第二步的结果直接转换成voc格式吗，如果不是将第二步结果转成voc格式第二步还有用吗？

为什么MCNN里面数据集标注格式是.mat 生成的密度图是.npy 但是到DMNet 标注格式就要.txt格式的 不明白 这也跑不动啊

First of all thanks for sharing.As you mentioned in your paper, you compute σ by estimating the average scale for each object category and estimate σ by applying Eq. 3

But I don't seem to see this point in this code, can you answer it for me, can you tell me which part of the code shows this part?

First of all, thank you for your selfless sharing
Can you please tell me how I can get the ground_truth of visdrone dataset, which is .mat file for remote sensing image, because I have a general understanding of MCNN, he uses the position of human head to build the .mat, how can I get it for visdrone?Because I see that both the K-nearest neighbor acquisition density map and the training process need this gt file, which is .mat
首先谢谢您无私的分享
麻烦问一下，对于使用MCNN生成密度图部分，对于遥感图像来说，如visdrone数据集，我该如何获得visdrone的ground_truth，也就是.mat文件，因为我大致了解了一下MCNN，他是用人头的位置构建的.mat，那对于visdrone来说我如何获得呢。因为我看到无论是K近邻获取密度图还是训练过程都需要这个gt文件，也就是.mat

Hello,thanks for your work.You provided the pretrained weight of the detector, but I don't find the train/test code of the detector, I know that is mmdetection, but how you train with the global data and crop data together? So I want the code that you change the config file, or the test code of the detector. I just want to run some images using your pretrained weight.Thank you very much.

Hello, the author, I would like to ask gaussian_kernels.pkl and distances_dict.pkl where there were published？What is the .param file I downloaded and how to use it? thank you!

VisDrone Crowd Counting dataset Ground Truth is in xml format, do you means to convert it to mat format and then send it to MCNN network to generate density map？ how to generate density map for VisDrones object detection dataset, its Ground Truth is different，please help me，thanks.

Hi,
There are some problems in this project:
1- in vision drone dataset's images are different size but density_slide_window.py request image size.
2- for image cropping, should be density map. this is not available reel time detection. it should be generate density map and image cropping every image in real time detection.
3- MCNN project generates density map but density map size smaller than image, this is not implemented in DMNet.
4- How am i using object detection model and where, i dont understand. Where detection model running in fusion_detection.py

Can you explain please?