search

mengyu212 / VIFNet_dehazing

An End-to-end Visible-Infrared Fusion Network for Image Dehazing
4 stars 0 forks source link

Image Resizing During Testing #2

Open s1mple-game opened 21 hours ago

s1mple-game commented 21 hours ago

In your code, it appears that the input image is resized to 640x480 during both training and testing. I followed this approach by resizing my input images to 640x480 during training the model, and the resulting performance closely matches the metrics reported in your paper.

However, when I resized the input images to 1920x1080 during testing, the performance significantly degraded.

I would like to know if there are any specific considerations or steps required when resizing the image during testing, especially when using a different resolution than the one used during training.

mengyu212 commented 18 hours ago

Thank you for your feedback. As you noted, our method is designed to work with a fixed image size of 640x480 for both training and testing. This fixed dimension helps ensure consistency in feature extraction and model performance. When resizing images to a different resolution, such as 1920x1080, it is likely that the model's ability to generalize diminishes, leading to the performance degradation you've observed. To maintain optimal performance, we recommend sticking with the specified input size of 640x480.