How to output the holistic score from the model?

[Max-WXB]

In the paper, you show exposure scores for images. But in the code you provided, I can only get a heatmap. I want to test the exposure of some images, how can I get the scores?

[xiewende]

In the paper, you show exposure scores for images. But in the code you provided, I can only get a heatmap. I want to test the exposure of some images, how can I get the scores?

the same quesion，do you have a solution now?

[Max-WXB]

In the paper, you show exposure scores for images. But in the code you provided, I can only get a heatmap. I want to test the exposure of some images, how can I get the scores?

the same quesion，do you have a solution now?

I used a simple method to estimate the score. I assumed that the green area in the heat map was the best exposure and set it to 10. The blue and red areas were the worst exposure and set to 0. Then I used a linear method to calculate the intermediate values ​​and finally calculated the average of all pixels.

[xiewende]

In the paper, you show exposure scores for images. But in the code you provided, I can only get a heatmap. I want to test the exposure of some images, how can I get the scores?

the same quesion，do you have a solution now?

I used a simple method to estimate the score. I assumed that the green area in the heat map was the best exposure and set it to 10. The blue and red areas were the worst exposure and set to 0. Then I used a linear method to calculate the intermediate values ​​and finally calculated the average of all pixels.

Good idea, could you provide me with your script at your convenience?

[Max-WXB]

In the paper, you show exposure scores for images. But in the code you provided, I can only get a heatmap. I want to test the exposure of some images, how can I get the scores?

the same quesion，do you have a solution now?

I used a simple method to estimate the score. I assumed that the green area in the heat map was the best exposure and set it to 10. The blue and red areas were the worst exposure and set to 0. Then I used a linear method to calculate the intermediate values ​​and finally calculated the average of all pixels.

Good idea, could you provide me with your script at your convenience?

I'm sorry I can't give you any code, this is a very easy to implement idea. You can use chatgpt or other ai to help you implement the code.

[xiewende]

In the paper, you show exposure scores for images. But in the code you provided, I can only get a heatmap. I want to test the exposure of some images, how can I get the scores?

the same quesion，do you have a solution now?

I used a simple method to estimate the score. I assumed that the green area in the heat map was the best exposure and set it to 10. The blue and red areas were the worst exposure and set to 0. Then I used a linear method to calculate the intermediate values ​​and finally calculated the average of all pixels.

Good idea, could you provide me with your script at your convenience?

I'm sorry I can't give you any code, this is a very easy to implement idea. You can use chatgpt or other ai to help you implement the code.

thanks，However, can this method distinguish between underexposure and overexposure?

[Max-WXB]

In the paper, you show exposure scores for images. But in the code you provided, I can only get a heatmap. I want to test the exposure of some images, how can I get the scores?

the same quesion，do you have a solution now?

I used a simple method to estimate the score. I assumed that the green area in the heat map was the best exposure and set it to 10. The blue and red areas were the worst exposure and set to 0. Then I used a linear method to calculate the intermediate values ​​and finally calculated the average of all pixels.

Good idea, could you provide me with your script at your convenience?

I'm sorry I can't give you any code, this is a very easy to implement idea. You can use chatgpt or other ai to help you implement the code.

thanks，However, can this method distinguish between underexposure and overexposure?

Obviously, this is a simple method, which can get a simple estimate of the exposure quality, but it seems that it is not very accurate for complex images because it calculates the average exposure quality of all pixels. This is easy to understand, it confuses overexposure and underexposure, so it cannot correctly give a correct quality. When you read the original code, you can find that they have built a MLP layer code, but for some reason, the training weights of this mlp layer are not given on github. I think rebuilding an MLP layer to regress the features is the most accurate solution.

[Max-WXB]

In the paper, you show exposure scores for images. But in the code you provided, I can only get a heatmap. I want to test the exposure of some images, how can I get the scores?

the same quesion，do you have a solution now?

I used a simple method to estimate the score. I assumed that the green area in the heat map was the best exposure and set it to 10. The blue and red areas were the worst exposure and set to 0. Then I used a linear method to calculate the intermediate values ​​and finally calculated the average of all pixels.

Good idea, could you provide me with your script at your convenience?

I'm sorry I can't give you any code, this is a very easy to implement idea. You can use chatgpt or other ai to help you implement the code.

thanks，However, can this method distinguish between underexposure and overexposure?

There is a simple method, you set the red area to -5 points, the blue area to 5 points, and then calculate their average, you can estimate whether the image is underexposed or overexposed. But I have not verified the accuracy of this method, it should be possible in theory.

[xiewende]

In the paper, you show exposure scores for images. But in the code you provided, I can only get a heatmap. I want to test the exposure of some images, how can I get the scores?

the same quesion，do you have a solution now?

I used a simple method to estimate the score. I assumed that the green area in the heat map was the best exposure and set it to 10. The blue and red areas were the worst exposure and set to 0. Then I used a linear method to calculate the intermediate values ​​and finally calculated the average of all pixels.

Good idea, could you provide me with your script at your convenience?

I'm sorry I can't give you any code, this is a very easy to implement idea. You can use chatgpt or other ai to help you implement the code.

thanks，However, can this method distinguish between underexposure and overexposure?

There is a simple method, you set the red area to -5 points, the blue area to 5 points, and then calculate their average, you can estimate whether the image is underexposed or overexposed. But I have not verified the accuracy of this method, it should be possible in theory.

I roughly looked at the prediction target and found that it is the difference between the normal exposure image and the test image. At this time, whether it is overexposure or underexposure, it is represented by the same distribution. It seems that simply setting the value cannot distinguish them.

[Max-WXB]

I roughly looked at the prediction target and found that it is the difference between the normal exposure image and the test image. At this time, whether it is overexposure or underexposure, it is represented by the same distribution. It seems that simply setting the value cannot distinguish them.

What I understand is that if we assume that the green area in the image is 0, the red area is -5, and the blue area is +5, then calculate its average score. If the average score is lower than 0, it is considered underexposed, and if the average score is higher than 0, it is considered overexposed. Of course, if an image has equal areas of underexposed and overexposed areas, the results will be averaged.

[xiewende]

I roughly looked at the prediction target and found that it is the difference between the normal exposure image and the test image. At this time, whether it is overexposure or underexposure, it is represented by the same distribution. It seems that simply setting the value cannot distinguish them.

What I understand is that if we assume that the green area in the image is 0, the red area is -5, and the blue area is +5, then calculate its average score. If the average score is lower than 0, it is considered underexposed, and if the average score is higher than 0, it is considered overexposed. Of course, if an image has equal areas of underexposed and overexposed areas, the results will be averaged.

My understanding is this: for example, if the difference between a pixel and a normal pixel is 0.2, how can we tell whether this pixel is overexposed or underexposed?
Or is it that when calculating the difference, positive numbers are used for overexposure and negative numbers are used for underexposure? If so, it is just as you said.

[Max-WXB]

My understanding is this: for example, if the difference between a pixel and a normal pixel is 0.2, how can we tell whether this pixel is overexposed or underexposed?
Or is it that when calculating the difference, positive numbers are used for overexposure and negative numbers are used for underexposure? If so, it is just as you said.

What I mean is to reversely infer whether it is overexposed or underexposed from the heat map. The heat map can already reflect the overexposed area and the underexposed area. Setting a threshold of -5 to 5 can simply calculate the proportion of overexposed area and underexposed area. I am not specialized in this research, so what I said may not be accurate.