The paper addresses the challenge of removing rain streaks from single images (de-raining) by introducing a recurrent framework that handles rain streaks of varying densities and shapes.

The network focuses on learning the "rain residual" (the difference between the rainy image and the clean background), making the training process more stable and effective. Content of the .rar File

Based on common distribution formats for this project, the DIDRPG2EMTL_comp.rar (or similar "comp" archives) typically contains:

.pth or .ckpt files that allow users to run the de-rain algorithm without training from scratch.

Settings for hyperparameters and directory paths used during the "comp" (computation/comparison) phase of the research. Performance and Impact

Instead of attempting to remove all rain in a single step, the model decomposes the rain layer into multiple stages. It progressively removes rain streaks by grouping them based on their physical characteristics.

The architecture uses recurrence to reuse parameters across different stages of the de-raining process, which reduces the model size while improving its ability to handle complex rain patterns.

Python implementation (often using PyTorch or TensorFlow).