RK3588 yolo最全总结

1.尽量选择relu激活函数

2.原forward函数记得修改为如下，再export为ONNX。

# 在models/yolo.py中，修改类Detect的forward函数

# def forward(self, x):
#     z = []  # inference output
#     for i in range(self.nl):
#         x[i] = self.m[i](x[i])  # conv
#         bs, _, ny, nx = x[i].shape  # x(bs,255,20,20) to x(bs,3,20,20,85)
#         x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()

#         if not self.training:  # inference
#             if self.dynamic or self.grid[i].shape[2:4] != x[i].shape[2:4]:
#                 self.grid[i], self.anchor_grid[i] = self._make_grid(nx, ny, i)

#             if isinstance(self, Segment):  # (boxes + masks)
#                 xy, wh, conf, mask = x[i].split((2, 2, self.nc + 1, self.no - self.nc - 5), 4)
#                 xy = (xy.sigmoid() * 2 + self.grid[i]) * self.stride[i]  # xy
#                 wh = (wh.sigmoid() * 2) ** 2 * self.anchor_grid[i]  # wh
#                 y = torch.cat((xy, wh, conf.sigmoid(), mask), 4)
#             else:  # Detect (boxes only)
#                 xy, wh, conf = x[i].sigmoid().split((2, 2, self.nc + 1), 4)
#                 xy = (xy * 2 + self.grid[i]) * self.stride[i]  # xy
#                 wh = (wh * 2) ** 2 * self.anchor_grid[i]  # wh
#                 y = torch.cat((xy, wh, conf), 4)
#             z.append(y.view(bs, self.na * nx * ny, self.no))

#     return x if self.training else (torch.cat(z, 1), ) if self.export else (torch.cat(z, 1), x)

def forward(self, x):
    z = []  # inference output
    for i in range(self.nl):
        z.append(torch.sigmoid(self.m[i](x[i])))

    return z

可选修改（不修改也能够成功跑通，但亦有手册如此建议，请自行查看，博主没有修改）

# 在export.py文件run()函数中修改：

    if half and not coreml:
        im, model = im.half(), model.half()  # to FP16
-   shape = tuple((y[0] if isinstance(y, tuple) else y).shape)  # model output shape
+   shape = tuple((y[0] if (isinstance(y, tuple) or (isinstance(y, list))) else y).shape)  # model output shape
    metadata = {'stride': int(max(model.stride)), 'names': model.names}  # model metadata
    LOGGER.info(f"\n{colorstr('PyTorch:')} starting from {file} with output shape {shape} ({file_size(file):.1f} MB)")

opset设置11/12 高版本容易出错

3.转换建议在x86上进行，使用镜像。板端建议使用RKNN Toolkit Lite。

4.尽管rk官方有提供函数rknn_lite.init_runtime(core_mask=RKNNLite.NPU_CORE_0_1_2)，使得RKNN模型能使用三核心进行推演，但NPU的使用率仍处于一个较低的水平。应考虑多线程/多进程方式，构建线程池，将视频帧绑定到核心。rknn_lite.init_runtime(core_mask=RKNNLite.NPU_CORE_0)
rknn_lite.init_runtime(core_mask=RKNNLite.NPU_CORE_1)
rknn_lite.init_runtime(core_mask=RKNNLite.NPU_CORE_2)

能够充分利用资源。此外应将其作定频操作，使其更加稳定发挥。

q2500050191
b站熬夜冲浪冠军
闲鱼认真专注开卷
Tb深度学习地表最强小店