pt --> TorchScript --> pnnx --> ncnnOptimize --> ncnn
import os
import torch
# 0. pt模型下载及初始化
model = torch.hub.load("PeterL1n/RobustVideoMatting", "mobilenetv3") # or "resnet50"
model.cpu()
model.eval()
# model(src, *rec, downsample_ratio=0.25) # src can be [B, C, H, W] or [B, T, C, H, W] RGB input is normalized to 0~1 range.
# 1. pt --> torchscript
traced_script_module = torch.jit.trace(model, torch.randn(1, 3, 512, 512), strict=False)
traced_script_module.save("ts.pt")
# 2. ts --> pnnx --> ncnn
os.system("pnnx ts.pt inputshape=[1,3,512,512],[1,3,320,320] device=cpu") # 可能错误
# 3. ncnn ---> optmize ----> ncnn
os.system("ncnnoptimize ts.ncnn.param ts.ncnn.bin opt.param opt.bin 1") # 数字0 代表fp32 ；1代表fp16报错内容如下：
# 运行后c++报错：原因暂时不清楚
# pass_level5
# pass_ncnn
# ignore pnnx.Expression pnnx_expr_256 param expr=1
# create_custom_layer pnnx.Expression
# fuse_convolution_activation conv_12 hswish_83
c++运行后报错：
layer pnnx_expr_256 not exists or registered