718c41634f
1.项目后端整体迁移至PaddleOCR-NCNN算法，已通过基本的兼容性测试 2.工程改为使用CMake组织，后续为了更好地兼容第三方库，不再提供QMake工程 3.重整权利声明文件，重整代码工程，确保最小化侵权风险 Log: 切换后端至PaddleOCR-NCNN，切换工程为CMake Change-Id: I4d5d2c5d37505a4a24b389b1a4c5d12f17bfa38c |
||
|---|---|---|
| .. | ||
| CMakeLists.txt | ||
| darknet2ncnn.cpp | ||
| output.jpg | ||
| README.md | ||
Darknet To NCNN Conversion Tools
This is a standalone darknet2ncnn converter without additional dependency.
Support yolov4, yolov4-tiny, yolov3, yolov3-tiny and enet-coco.cfg (EfficientNetB0-Yolov3).
Another conversion tool based on darknet can be found at: darknet2ncnn
Usage
Usage: darknet2ncnn [darknetcfg] [darknetweights] [ncnnparam] [ncnnbin] [merge_output]
[darknetcfg] .cfg file of input darknet model.
[darknetweights] .weights file of input darknet model.
[cnnparam] .param file of output ncnn model.
[ncnnbin] .bin file of output ncnn model.
[merge_output] merge all output yolo layers into one, enabled by default.
Example
1. Convert yolov4-tiny cfg and weights
Download pre-trained yolov4-tiny.cfg and yolov4-tiny.weights or with your own trained weight.
Convert cfg and weights:
./darknet2ncnn yolov4-tiny.cfg yolov4-tiny.weights yolov4-tiny.param yolov4-tiny.bin 1
If succeeded, the output would be:
Loading cfg...
WARNING: The ignore_thresh=0.700000 of yolo0 is too high. An alternative value 0.25 is written instead.
WARNING: The ignore_thresh=0.700000 of yolo1 is too high. An alternative value 0.25 is written instead.
Loading weights...
Converting model...
83 layers, 91 blobs generated.
NOTE: The input of darknet uses: mean_vals=0 and norm_vals=1/255.f.
NOTE: Remember to use ncnnoptimize for better performance.
2. Optimize graphic
./ncnnoptimize yolov4-tiny.param yolov4-tiny.bin yolov4-tiny-opt.param yolov4-tiny-opt.bin 0
3. Test
build examples/yolov4.cpp and test with:
./yolov4 dog.jpg
The result will be:
How to run with benchncnn
Set 2=0.3 for Yolov3DetectionOutput layer.