DehazeXLv2

遥感图像去云雾权重。Remote-sensing image dehazing weights.

中文

项 内容
文件 dehazexlv2_RS_v0_0.pth
参数量 36.20M
大小 约 145MB
配置 dehazexlv2_default.yaml,必须和这份权重一起用
输入 RGB,像素缩放到 [0, 1]

文件里只有模型参数,不含优化器和训练状态。

下载:

huggingface-cli download fengyanzi/Dehazexlv2 dehazexlv2_RS_v0_0.pth --local-dir weights
huggingface-cli download fengyanzi/Dehazexlv2 dehazexlv2_default.yaml --local-dir configs

推理:

from dehazexlv2 import DehazeXL

model = DehazeXL("weights/dehazexlv2_RS_v0_0.pth", config="configs/dehazexlv2_default.yaml")
model("image.png")

不指定设备时自动选择:没有 GPU 用 CPU,一张卡用这张卡,多张卡时多张图按卡并行。

English

Item Value
File dehazexlv2_RS_v0_0.pth
Parameters 36.20M
Size about 145MB
Config dehazexlv2_default.yaml, required with these weights
Input RGB, pixels scaled to [0, 1]

The file contains model parameters only. Optimizer state and training metadata are not included.

Download:

huggingface-cli download fengyanzi/Dehazexlv2 dehazexlv2_RS_v0_0.pth --local-dir weights
huggingface-cli download fengyanzi/Dehazexlv2 dehazexlv2_default.yaml --local-dir configs

Inference:

from dehazexlv2 import DehazeXL

model = DehazeXL("weights/dehazexlv2_RS_v0_0.pth", config="configs/dehazexlv2_default.yaml")
model("image.png")

With no device set, the loader checks the machine: CPU when no GPU is present, the single GPU when one is present, and one image per GPU when several images are restored on several GPUs.

The network matches the config: channels 80, 128, 256, 512, with tile / halo of 256 / 32.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Space using fengyanzi/Dehazexlv2 1