--- tags: NTU-IoT-Cow, yolo --- # YOLOv4-Tiny Experiments [Reference](https://github.com/AlexeyAB/darknet) ## Calculate anchor box with different ratio ```bash= ./darknet detector calc_anchors \ data/obj.data \ -num_of_clusters 7 \ -width 416 -height 416 ``` edit `.cfg` corresponding to the numbers of anchor boxes ## Train - Train and log ```bash= ./darknet detector train \ data/obj.data \ obj.cfg \ yolov4-tiny.conv.29 -map -gpus 0,1 | tee ./log/obj.log ``` ## Validate - Validate weight ```bash= ./darknet detector map \ ./data/obj.data \ ./obj.cfg \ ./1000.weight ``` - Run validation ```bash= ./darknet detector valid \ data/obj.data \ obj.cfg \ obj.weights \ -thresh 0.5 ``` - Predict a batch of file ```bash= ./darknet detector test \ cfg/obj.data \ obj.cfg \ yolov4.weights \ -ext_output -dont_show \ -out result.json < ./data/train.txt ``` `train.txt` contain the list of images which you want to predict. ###### tags: `yolo`, `tutorial`