ImageDehazing

This is a collection of image dehazing algorithms, models have been converted to run on the Axera NPU using w8a8 quantization.

This model has been optimized with the following LoRA:

Compatible with Pulsar2 version: 6.0 115775d3

Convert tools links:

For those who are interested in model conversion, you can try to export axmodel through

• Pulsar2 Link, How to Convert ONNX to axmodel

Support Platform

• AX650
  • M4N-Dock(爱芯派Pro)
  • M.2 Accelerator card

模型	输入分辨率	AX650板端耗时
AOD-Net	640×480	2.4 ms
Light-Dehazenet	480×640	9.2 ms
DehazeFormer_t	512×512	161ms
MixDehazeNet	256×256	38 ms
GCANet	512×512	80 ms
GridDehazeNet	640×480	113 ms
DEA-Net	512×512	127 ms
FFA-Net	512×512	873 ms

How to use

Download all files from this repository to the device

所有模型文件组织方式均如下：
.
|-- model_convert
|   |-- axmodel
|   |   `-- dehazeformer-t-512-constant.axmodel
|   `-- dehazeformer.json
|-- pic
|   `-- 00000_0_0.1800.png
|-- python
|   |-- axmodel_infer.py
|   `-- onnx_infer.py
`-- res
    `-- output.png

Inference

Inference with AX650 Host, such as M4N-Dock(爱芯派Pro)

模型推理，执行命令:

(base) root@ax650:~/GCANet# python axmodel_infer.py
[INFO] Available providers:  ['AxEngineExecutionProvider', 'AXCLRTExecutionProvider']
[INFO] Using provider: AxEngineExecutionProvider
[INFO] Chip type: ChipType.MC50
[INFO] VNPU type: VNPUType.DISABLED
[INFO] Engine version: 2.12.0s
[INFO] Model type: 2 (triple core)
[INFO] Compiler version: 7.0 22923e4e
Saved: axmodel_output/0051_0.8_0.2_input_dehaze_compare.png
Saved: axmodel_output/0099_0.9_0.16_input_dehaze_compare.png

推理结果样例：