DehazeXLv2
weights for image haze removal
1 min readupdated
A specialized weight file dehazexlv2_RS_v0_0.pth for removing fog, haze, and clouds from remote sensing (satellite and aerial) images. The model is built on a compact architecture with 36.20M parameters (weighing about 145MB) and requires the dehazexlv2_default.yaml configuration file to run. It takes RGB images as input with pixels normalized to the [0, 1] range. This tool will come in handy for researchers, surveyors, and developers working with satellite imagery processing and Earth observation data.
Key features
- Compact weight file size (about 145MB) with 36.20M parameters.
- Requires the
dehazexlv2_default.yamlconfiguration file to operate. - Automatic device selection (CPU, single GPU, or multi-GPU parallel processing).
- Architecture with channel counts of 80, 128, 256, 512 and a tile/halo size of 256 / 32.
What you can do
- Improve visibility and contrast in hazy satellite or aerial photos.
- Integrate processing into geographic information system pipelines with support for multi-threaded execution across multiple GPUs.
Example
Running model inference for image restoration:
from dehazexlv2 import DehazeXL
model = DehazeXL("weights/dehazexlv2_RS_v0_0.pth", config="configs/dehazexlv2_default.yaml")
model("image.png")Read more
DetailsFile:
dehazexlv2_RS_v0_0.pthParameters: 36.20M, size ~145MB
Input data: RGB, pixels in the [0, 1] range
Model weights only (optimizer state excluded)










Comments