通过下载yolo6的程序
yOLOv6 是美团视觉智能部研发的一款目标检测框架,致力于工业应用。本框架同时专注于检测的精度和推理效率
YOLOv6 主要在 BackBone、Neck、Head 以及训练策略等方面进行了诸多的改进:
- 统一设计了更高效的 Backbone 和 Neck :受到硬件感知神经网络设计思想的启发,基于 RepVGG style[4] 设计了可重参数化、更高效的骨干网络 EfficientRep Backbone 和 Rep-PAN Neck。
- 优化设计了更简洁有效的 Efficient Decoupled Head,在维持精度的同时,进一步降低了一般解耦头带来的额外延时开销。
- 在训练策略上,我们采用Anchor-free 无锚范式,同时辅以 SimOTA[2] 标签分配策略以及 SIoU[9] 边界框回归损失来进一步提高检测精度。
然后进行测试
用了ylov6 s的权重文件 进行测试
可以看出 6s的效果是最好的
得到的了结果如上所示
python tools/infer.py –weights yolov6s.pt –source image1.jpg
python tools/infer.py –weights yolov6s.pt –source data/images 代表了测试文件夹下面的材料
YOLOv6传送门:https://github.com/meituan/YOLOv6
在这里
经过一会 咱们在看训练的效果如何
2/399 3.002 1.709 5.856 1.522: 40%|████ | 17/42 [00:36<00:51, 2.08s/it] 2/399 3.002 1.709 5.856 1.522: 43%|████▎ | 18/42 [00:36<00:48, 2.02s/it] 2/399 2.999 1.708 5.858 1.526: 43%|████▎ | 18/42 [00:38<00:48, 2.02s/it] 2/399 2.999 1.708 5.858 1.526: 45%|████▌ | 19/42 [00:38<00:46, 2.02s/it] 2/399 2.994 1.707 5.839 1.526: 45%|████▌ | 19/42 [00:40<00:46, 2.02s/it] 2/399 2.994 1.707 5.839 1.526: 48%|████▊ | 20/42 [00:40<00:43, 1.98s/it] 2/399 2.989 1.706 5.814 1.527: 48%|████▊ | 20/42 [00:42<00:43, 1.98s/it] 2/399 2.989 1.706 5.814 1.527: 50%|█████ | 21/42 [00:42<00:41, 1.97s/it] 2/399 2.989 1.703 5.828 1.526: 50%|█████ | 21/42 [00:44<00:41, 1.97s/it] 2/399 2.989 1.703 5.828 1.526: 52%|█████▏ | 22/42 [00:44<00:39, 1.96s/it] 2/399 2.982 1.701 5.84 1.534: 52%|█████▏ | 22/42 [00:46<00:39, 1.96s/it] 2/399 2.982 1.701 5.84 1.534: 55%|█████▍ | 23/42 [00:46<00:37, 1.95s/it] 2/399 2.982 1.699 5.836 1.53: 55%|█████▍ | 23/42 [00:48<00:37, 1.95s/it] 2/399 2.982 1.699 5.836 1.53: 57%|█████▋ | 24/42 [00:48<00:35, 1.95s/it] 2/399 2.979 1.696 5.816 1.531: 57%|█████▋ | 24/42 [00:50<00:35, 1.95s/it] 2/399 2.979 1.696 5.816 1.531: 60%|█████▉ | 25/42 [00:50<00:32, 1.92s/it] 2/399 2.974 1.695 5.798 1.531: 60%|█████▉ | 25/42 [00:52<00:32, 1.92s/it] 2/399 2.974 1.695 5.798 1.531: 62%|██████▏ | 26/42 [00:52<00:30, 1.93s/it] 2/399 2.976 1.696 5.786 1.53: 62%|██████▏ | 26/42 [00:54<00:30, 1.93s/it] 2/399 2.976 1.696 5.786 1.53: 64%|██████▍ | 27/42 [00:54<00:28, 1.93s/it] 2/399 2.978 1.696 5.777 1.526: 64%|██████▍ | 27/42 [00:56<00:28, 1.93s/it] 2/399 2.978 1.696 5.777 1.526: 67%|██████▋ | 28/42 [00:56<00:26, 1.93s/it] 2/399 2.976 1.692 5.776 1.528: 67%|██████▋ | 28/42 [00:58<00:26, 1.93s/it]
目前这是我训练的操作过程