Merge pull request #19 from Laughing-q/instance_seg

fix accuracy issue

models/yolo.py

@@ -135,7 +135,7 @@ def forward(self, x):
                     y[..., 0:2] = (y[..., 0:2] * 2 + self.grid[i]) * self.stride[i]  # xy
                     y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i]  # wh
                 else:  # for YOLOv5 on AWS Inferentia https://github.com/ultralytics/yolov5/pull/2953
-                    xy = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i]  # xy
+                    xy = (y[..., 0:2] * 2. + self.grid[i]) * self.stride[i]  # xy
                     wh = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i]  # wh
                     y = torch.cat((xy.type_as(y), wh.type_as(y), y[..., 4:]), -1)
                 z.append(y.view(-1, self.na * ny * nx, self.no))

segment/predict.py

@@ -164,7 +164,7 @@ def run(
                     s += f"{n} {names[int(c)]}{'s' * (n > 1)}, "  # add to string
 
                 # Mask plotting ----------------------------------------------------------------------------------------
-                mcolors = [colors(int(cls)) for cls in range(len(det[:, 5]))]
+                mcolors = [colors(int(cls), True) for cls in det[:, 5]]
                 # NOTE: this plot method is faster, but the image might get blurred https://github.com/dbolya/yolact
                 img_masks = plot_masks(im[i], masks, mcolors)  # image with masks shape(imh,imw,3)
                 img_masks = scale_masks(im.shape[2:], img_masks, im0.shape)  # scale to original h, w

utils/general.py

@@ -900,6 +900,8 @@ def non_max_suppression(
             continue
         elif n > max_nms:  # excess boxes
             x = x[x[:, 4].argsort(descending=True)[:max_nms]]  # sort by confidence
+        else:
+            x = x[x[:, 4].argsort(descending=True)]  # sort by confidence
 
         # Batched NMS
         c = x[:, 5:6] * (0 if agnostic else max_wh)  # classes