chenchao666 / bone-age-assessment Goto Github PK

I generated heatmap for one region but there is no code for other regions like in aggregation.py file we are using dataR2.pkl. dataR1.pkl dataHand.pkl but there is no code where we have pickled these files before

@chenchao666 thanks for your working and sharing! when I implement the code as your repository, there are some differentiations as your paper, could you helpe me?

1. when i implement the code your repository, i get the AttentionImg and heatmep as your repository too, is this like the R-1 region? But in the paper, R-1 should be resized to 560 * 560.
2. when i resize the original fig to 560 * 560, the output size of base_model is 16 * 16 * 2048, your paper is 18 * 18 * 2048, could you tell me where i have mistakes?
3. for get the Hand-region, i resized the fig to 300 * 300, and changed the thresehold to 20, but i got the attention-region was small as R-1, can't crop the whole hand from original image.
4. in the attention-image step, the gender infomation is not use, right?
   looking forward to your reply, thanks again.
   sincerely, yours!

What are you doing here?

The gender should be one feature vector with 1 for male and 0 for female, isn't it? It causes gradients to exploit or vanishing gradients in training time for such inputs.

heatmap而不是heatmep

请问博主有pytorch的版本吗？谢谢

Is it normal for the training of the classification model to take more than 10 hours per epoch during the second step??
specific properties for my device:
processor :ADM Ryzen 5 4600 with Radeon Graphics 3.00 Ghz
RAM: 24 GB
System type:64-bit operating system
NVIDIA GeForce GTX 1660 Ti graphics card

As the title, I'm a bit confused about why the localization network is classification, why not regression? The final goal of the paper is regression, right? Can you explain it?
Many thanks.

about the Step 3: Generate the Heatmap for the Hand Reigon, Region-1 and Region-2 one-by-one, I want to know:
How to generate the Heatmap for the Hand Reigon, Region-1 and Region-2 one-by-one?