Support norm model

deploy/NCNN/README.md

@@ -39,11 +39,12 @@ python ./deploy/NCNN/export_torchscript.py \
 ## Run inference with NCNN-Python
 
 ```shell
-python3 deploy/NCNN/infer-lite-model.py \
+python3 deploy/NCNN/infer-ncnn-model.py \
   data/images/image1.jpg \
   work_dir/yolov6lite_s.ncnn.param \
   work_dir/yolov6lite_s.ncnn.bin \
   --img-size 320 320 \
+  --max-stride 64 \
   --show
 ```
 
@@ -57,6 +58,10 @@ python3 deploy/NCNN/infer-lite-model.py \
 - `--img-size` : The image height and width for model input.
 - `--max-stride` : The yolov6 lite model max stride.
 
+***Notice!***
+
+If you want to try norm yolov6 model such as `yolov6n/s/m/l`, you should add `--max-stride 32` flags .
+
 
 ## Download
 
@@ -65,3 +70,7 @@ python3 deploy/NCNN/infer-lite-model.py \
 * [YOLOv6-lite-l]()
 * [YOLOv6-lite-l-320x192]()
 * [YOLOv6-lite-l-224x128]()
+* [YOLOv6-n]()
+* [YOLOv6-s]()
+* [YOLOv6-m]()
+* [YOLOv6-l]()

deploy/NCNN/infer-lite-model.py → deploy/NCNN/infer-ncnn-model.py

@@ -71,6 +71,9 @@ def sigmoid(x: ndarray) -> ndarray:
                         (255, 149, 200), (255, 55, 199)],
                        dtype=np.uint8)
 
+CONF_THRES = 0.45
+IOU_THRES = 0.65
+
 
 def parse_args() -> argparse.Namespace:
     parser = argparse.ArgumentParser()
@@ -92,6 +95,7 @@ def parse_args() -> argparse.Namespace:
         default=64,
         help='Max stride of yolov6 model')
     args = parser.parse_args()
+    assert args.max_stride in (32, 64)
     return args
 
 
@@ -201,8 +205,8 @@ def main(args: argparse.Namespace):
         img, ncnn.Mat.PixelType.PIXEL_BGR2RGB, img_w, img_h, w, h
     )
 
-    wpad = (w + 63) // 64 * 64 - w
-    hpad = (h + 63) // 64 * 64 - h
+    wpad = (w + args.max_stride - 1) // args.max_stride * args.max_stride - w
+    hpad = (h + args.max_stride - 1) // args.max_stride * args.max_stride - h
 
     mat_in_pad = ncnn.copy_make_border(
         mat_in,
@@ -221,10 +225,14 @@ def main(args: argparse.Namespace):
     ret1, mat_out1 = ex.extract("out0")  # stride 8
     ret2, mat_out2 = ex.extract("out1")  # stride 16
     ret3, mat_out3 = ex.extract("out2")  # stride 32
-    ret4, mat_out4 = ex.extract("out3")  # stride 64
+    if args.max_stride == 64:
+        ret4, mat_out4 = ex.extract("out3")  # stride 64
+
+    outputs = [np.array(mat_out1), np.array(mat_out2), np.array(mat_out3)]
+    if args.max_stride == 64:
+        outputs.append(np.array(mat_out4))
 
-    outputs = [np.array(mat_out1), np.array(mat_out2), np.array(mat_out3), np.array(mat_out4)]
-    nmsd_boxes, nmsd_scores, nmsd_labels = yolov6_decode(outputs, 0.45, 0.65)
+    nmsd_boxes, nmsd_scores, nmsd_labels = yolov6_decode(outputs, CONF_THRES, IOU_THRES)
 
     for box, score, label in zip(nmsd_boxes, nmsd_scores, nmsd_labels):
         x0, y0, x1, y1 = box