README.md

metadata

pipeline_tag: image-to-image

SuperResolution

This version of SuperResolution has been converted to run on the Axera NPU using w8a8 quantization.

This model has been optimized with the following LoRA:

Compatible with Pulsar2 version: 4.2

Convert tools links:

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

• The github repo of AXera Platform

• How to Convert ONNX to axmodel

• Pulsar2 tools

Support Platform

• AX650
  • M4N-Dock(爱芯派Pro)
  • M.2 Accelerator card
• AX630C
  • 爱芯派2
  • Module-LLM
  • LLM630 Compute Kit

Chips	model	cost
AX650	EDSR	800 ms
	ESPCN	22 ms

How to use

Download all files from this repository to the device

root@ax650:~/SuperResolution# tree
.
|-- model_convert
|	-- axmodel
|   	`-- edsr_baseline_x2_1.axmodel
|   	`-- espcn_x2_T9.axmodel
|	-- onnx
|   	`-- edsr_baseline_x2_1.onnx
|   	`-- espcn_x2_T9.onnx
|   `-- build_config_edsr.json
|   `-- build_config_espcn.json
|-- python
|   `-- run_onnx.py
|   `-- run_axmodel.py
|   `-- common.py
|   `-- imgproc.py
|-- video
|   `-- test_1920x1080.mp4
|   `-- 1.png
|   `-- 2.png

Inference

Input Data:

|-- video
|   `-- test_1920x1080.mp4

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

root@ax650 ~/SuperResolution #python3 run_axmodel.py --model ./espcn_x2_T9.axmodel --scale 2 --dir_demo ./test_1920x1080.mp4
[INFO] Available providers:  ['AxEngineExecutionProvider']
[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: 4.2 6bff2f67
100%|█████████████████████████████████████████| 267/267 [10:06<00:00,  2.27s/it]
Total time: 99.582 seconds for 267 frames
Average time: 0.373 seconds for each frame

Output: