Darknet YOLO v3 and v4

• v4 Paper: [2004.10934] YOLOv4: Optimal Speed and Accuracy of Object Detection
• Repository: AlexeyAB/darknet: YOLOv4 - Neural Networks for Object Detection (Windows and Linux version of Darknet )
• Google Colab: YOLOv4_Tutorial.ipynb - Colaboratory

Build Docker Image

cd ./docs/Tutorial-YOLO-Darknet

Edit Makefile

利用する GPU ごとに Makefile を編集する必要がある。

GeForce RTX 2080 Ti であれば以下のようにする。(デフォルト)

# GeForce RTX 2080 Ti, RTX 2080, RTX 2070, Quadro RTX 8000, Quadro RTX 6000, Quadro RTX 5000, Tesla T4, XNOR Tensor Cores
ARCH= -gencode arch=compute_75,code=[sm_75,compute_75]

# GTX 1080, GTX 1070, GTX 1060, GTX 1050, GTX 1030, Titan Xp, Tesla P40, Tesla P4
# ARCH= -gencode arch=compute_61,code=sm_61 -gencode arch=compute_61,code=compute_61

GTX 1080 であれば以下のようにする。

# GeForce RTX 2080 Ti, RTX 2080, RTX 2070, Quadro RTX 8000, Quadro RTX 6000, Quadro RTX 5000, Tesla T4, XNOR Tensor Cores
# ARCH= -gencode arch=compute_75,code=[sm_75,compute_75]

# GTX 1080, GTX 1070, GTX 1060, GTX 1050, GTX 1030, Titan Xp, Tesla P40, Tesla P4
ARCH= -gencode arch=compute_61,code=sm_61 -gencode arch=compute_61,code=compute_61

Build

$ nvidia-smi
NVIDIA-SMI 440.82       Driver Version: 440.82       CUDA Version: 10.2

nvidia-smi で確認できる CUDA Version と Base Docker image nvidia/cuda の tag を一致させる。

docker build . -t darknet-yolo:latest --build-arg BASE_IMAGE="nvidia/cuda:10.2-cudnn7-devel-ubuntu18.04"

Run Container

docker run --rm -i -t --gpus all darknet-yolo:latest bash

Predict

推論の実行例を以下に示す。

$ ./darknet detector test cfg/coco.data cfg/yolov3.cfg yolov3.weights data/dog.jpg
...
data/dog.jpg: Predicted in 12.776000 milli-seconds.
bicycle: 99%
dog: 100%
truck: 94%
...

$ ./darknet detector test cfg/coco.data cfg/yolov4.cfg yolov4.weights data/dog.jpg
...
data/dog.jpg: Predicted in 24.482000 milli-seconds.
bicycle: 92%
dog: 98%
truck: 92%
pottedplant: 33%
...

$ python3 darknet.py
...
Total BFLOPS 128.459
avg_outputs = 1068395
 Allocate additional workspace_size = 52.43 MB
 Try to load weights: yolov4.weights
Loading weights from yolov4.weights...
 seen 64, trained: 32032 K-images (500 Kilo-batches_64)
Done! Loaded 162 layers from weights-file
Loaded - names_list: data/coco.names, classes = 80
Unable to show image: No module named 'skimage'
[('dog', 0.9787506461143494, (220.98822021484375, 383.2079772949219, 184.41787719726562, 316.5090637207031)), ('bicycle', 0.921798586845398, (343.4819641113281, 276.87603759765625, 458.0648193359375, 298.7120361328125)), ('truck', 0.91830974817276, (574.2606201171875, 123.24830627441406, 220.67361450195312, 93.20550537109375)), ('pottedplant', 0.33072134852409363, (699.3265380859375, 131.88845825195312, 36.533931732177734, 45.44673538208008))]

v3	v4