Hi, thank you for sharing your code! This model seems very interesting and promising. I was trying to test your model on a video, but unfortunately I was not able to load your pre-trained models. I tried both HR and VGG models, but it always break on load_state_dict() Do you know why?

netName = 'HR_Net'
GPU_ID = '0'
torch.backends.cudnn.benchmark = True
os.environ["CUDA_VISIBLE_DEVICES"] = GPU_ID
model_path = './saved_model/NWPU-HR-ep_241_F1_0.802_Pre_0.841_Rec_0.766_mae_55.6_mse_330.9.pth'
net = Crowd_locator(netName,GPU_ID,pretrained=True)
net.load_state_dict(torch.load(model_path))
net.eval()

File "/home/walter/anaconda3/envs/crowd/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1052, in load_state_dict self.class.name, "\n\t".join(error_msgs))) RuntimeError: Error(s) in loading state_dict for Crowd_locator: Missing key(s) in state_dict: "Extractor.conv1.weight", "Extractor.bn1.weight", "Extractor.bn1.bias", "Extractor.bn1.running_mean", "Extractor.bn1.running_var", "Extractor.conv2.weight", "Extractor.bn2.weight", "Extractor.bn2.bias", "Extractor.bn2.running_mean", "Extractor.bn2.running_var", "Extractor.layer1.0.conv1.weight", "Extractor.layer1.0.bn1.weight", "Extractor.layer1.0.bn1.bias", "Extractor.layer1.0.bn1.running_mean", "Extractor.layer1.0.bn1.running_var", "Extractor.layer1.0.conv2.weight", "Extractor.layer1.0.bn2.weight", "Extractor.layer1.0.bn2.bias", "Extractor.layer1.0.bn2.running_mean", [...]

Hello, your work is interesting and inspired. I'm trying to re-implement your paper using keras, but the weights of threshold encoder would be nan after some epochs. Have this phenomenon happened to you, or could you give some suggestions?

thank you~

It is interesting to consider the crowd localization as as segmentation task, impressive!

I wonder to know that have you compare your methods with other well known segmentation methods? It seems that common segmentation network can also be trained with the mask.

Also, in Table 3, it seems that the lower fixed value of threshold, the better performance will be got. Have you try thresholds lower than 0.5? As far as I think, if the center is the only thing needed, the lower threshold will have better performance.

By the way, your IBM/PBM module looks also suitable for other segmentation task, have you test it on other dataset such as COCO?

Best regards.

Ho guys,

How to fix this issue

RuntimeError: CUDA error: out of memory CUDA kernel errors might be asynchronously reported at some other API call,so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1.

您好，首先感谢您分享了您的稠密目标定位算法，最近我对您得算法进行实现得时候出现了问题，如下： 首先我是在windows10平台上运行得训练过程，前面数据和模型处理好了以后，开始训练出现问题如下： 不论运行那个模型，当模型开始正向执行，总是在第一个conv出爆出问题： VGG16_FPN.py: def forward(self, x): f = [] x = self.layer1(x) seg_hrnet.py: def forward(self, x): residual = x out = self.conv1(x) 总是爆出内存不足问题， RuntimeError: CUDA out of memory. Tried to allocate 4.50 GiB (GPU 0; 12.00 GiB total capacity; 886.66 MiB already allocated; 5.14 GiB free; 4.94 GiB reserved in total by PyTorch) 但是其中它想去分配4.5G，（Tried to allocate 4.50 GiB），而我的电脑除了pytorch占用得，还有5G(5.14 GiB free; 4.94 GiB reserved in total by PyTorch). 目前不知道是哪里除了问题，请问您原始代码是在那个平台训练得，ubuntu吗？或者您这边有什么思路可以解决此问题吗？ 期待您得回复，万分感谢！

Hi guys,

I want to train this model on(NVIDIA GeForce RTX 2080 Ti/PCIe/SSE2), RAM 16GB, How to config it. When I ran the default setting, the bellow error has occored.

    return t.to(device, dtype if t.is_floating_point() or t.is_complex() else None, non_blocking)
RuntimeError: CUDA error: out of memory
CUDA kernel errors might be asynchronously reported at some other API call,so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.

Hi tao han: I am a graduate student in SEU, trying to replace the backbone of IIM (VGG16_FPN or HRNet) to my Transformer crowd counting model. However, even I low the initial lr 2 1e-6 to 1e-7 in SHA. the threshold even appears to be NAN in the 700 epoch. Also, the best MAE is only 126, which is far away from my model combined with other losses (more than MSE) on SHA. I noticed that in this link https://github.com/taohan10200/IIM/issues/7#issuecomment-766274210 you have mentioned that we also could lower the initial threshold, I wonder to sure whether is the initial weight 0.5 in the Binarized module. But even I change the initial weight to 0.4, the t_max also starts with 0.54. I get confused with the Binarized module. looking forward to your reply, my email is [email protected]/ [email protected]

> class BinarizedModule(nn.Module):
>   def __init__(self, input_channels=720):
>     super(BinarizedModule, self).__init__()
>     self.Threshold_Module = nn.Sequential(
>         nn.Conv2d(input_channels, 256, kernel_size=3, stride=1, padding=1, bias=False),
> adding=0, bias=False),
>         nn.AvgPool2d(15, stride=1, padding=7),
>     )
>     self.sig = compressedSigmoid()
>     #Change the threshold org to 0.4
>     self.weight = nn.Parameter(torch.Tensor(1).fill_(0.4),requires_grad=True)
>     self.bias = nn.Parameter(torch.Tensor(1).fill_(0), requires_grad=True)

Thanks for the author's impressive work! Question 1: How to visualize test set pictures for SHHA, SHHB, UCF-QNRF, FDST?

Question 2: The link (DATA.md) is broken, could you please update the link, especially SHHA and SHHB, thanks.

Thanks for the author's work.Great work!!!!! 1:Working so hard to train the Threshold Encoder which aim is to seperate the pred_map(the output of the network) to 1 or 0, but there is little differences between the pred_map after processed by the Threshold Encoder and the pred_map without processed by the Threshold Encoder.Details can look at the picture

您好，在尝试使用torch.jit.trace保存为pt模型文件时 traced_script_module = torch.jit.trace(net,torch.rand(1, 3, 224, 224)) traced_script_module.save("modelIIMX.pt") 报如下错误： Traceback (most recent call last): File "gpt.py", line 28, in torch.jit.save(model,"modelIIMX.pt") File "/home/wjq/anaconda3/envs/IIM/lib/python3.7/site-packages/torch/jit/_serialization.py", line 81, in save m.save(f, _extra_files=_extra_files) File "/home/wjq/anaconda3/envs/IIM/lib/python3.7/site-packages/torch/jit/_script.py", line 490, in save return self._c.save(str(f), **kwargs) RuntimeError: Could not export Python function call 'BinarizedF'. Remove calls to Python functions before export. Did you forget to add @script or @script_method annotation? If this is a nn.ModuleList, add it to constants: /home/wjq/IIM/IIM-main/model/PBM.py(77): forward /home/wjq/anaconda3/envs/IIM/lib/python3.7/site-packages/torch/nn/modules/module.py(860): _slow_forward /home/wjq/anaconda3/envs/IIM/lib/python3.7/site-packages/torch/nn/modules/module.py(887): _call_impl /home/wjq/IIM/IIM-main/model/locator.py(39): forward /home/wjq/anaconda3/envs/IIM/lib/python3.7/site-packages/torch/nn/modules/module.py(860): _slow_forward /home/wjq/anaconda3/envs/IIM/lib/python3.7/site-packages/torch/nn/modules/module.py(887): _call_impl /home/wjq/anaconda3/envs/IIM/lib/python3.7/site-packages/torch/jit/_trace.py(940): trace_module /home/wjq/anaconda3/envs/IIM/lib/python3.7/site-packages/torch/jit/_trace.py(742): trace gpt.py(27): 尝试了一些方法仍未解决，麻烦有空帮忙看下要如何解决，谢谢