高端定制网站开发建站教程详解网站设置了权限
目录
一、理由
二、方法
（1）导入MobileViTv3模块
（2）在ultralytics/nn/tasks.py的函数parse_model中修改
（3）在yaml配置文件中写入
（4）开始训练，先把其他梯度关闭，保留新加的模块的梯度。
代码已在GitHub上传，链接：yolov8_vit
一、理由
MobileViTv3是一种为移动设备优化的轻量级视觉Transformer架构，它结合了卷积神经网络（CNN）和视觉Transformer（ViT）的特点，以创建适合移动视觉任务的轻量级模型。
二、方法
（1）导入MobileViTv3模块
在ultralytics/nn创建vit文件夹，文件夹内放MobileViTv3以及需要的包。MobileViTv3模块如下：
import numpy as np
from torch import nn, Tensor
import math
import torch
from torch.nn import functional as F
from typing import Optional, Dict, Tuple, Union, Sequence
from mobilevit_v2_block import MobileViTBlockv2 as MbViTBkV2class MbViTV3(MbViTBkV2):def __init__(self,in_channels: int,attn_unit_dim: int,patch_h: Optional[int] = 2,patch_w: Optional[int] = 2,ffn_multiplier: Optional[Union[Sequence[Union[int, float]], int, float]] = 2.0,n_attn_blocks: Optional[int] = 2,attn_dropout: Optional[float] = 0.0,dropout: Optional[float] = 0.0,ffn_dropout: Optional[float] = 0.0,conv_ksize: Optional[int] = 3,attn_norm_layer: Optional[str] = "layer_norm_2d",enable_coreml_compatible_fn: Optional[bool] = False,) -> None:super(MbViTV3, self).__init__(in_channels, attn_unit_dim)self.enable_coreml_compatible_fn = enable_coreml_compatible_fnif self.enable_coreml_compatible_fn:# we set persistent to false so that these weights are not part of model's state_dictself.register_buffer(name="unfolding_weights",tensor=self._compute_unfolding_weights(),persistent=False,)cnn_out_dim = attn_unit_dimself.conv_proj = nn.Conv2d(2 * cnn_out_dim, in_channels, 1, 1)def forward_spatial(self, x: Tensor, *args, **kwargs) -> Tensor:x = self.resize_input_if_needed(x)fm_conv = self.local_rep(x)# convert feature map to patchesif self.enable_coreml_compatible_fn:patches, output_size = self.unfolding_coreml(fm_conv)else:patches, output_size = self.unfolding_pytorch(fm_conv)# learn global representations on all patchespatches = self.global_rep(patches)# [B x Patch x Patches x C] --> [B x C x Patches x Patch]if self.enable_coreml_compatible_fn:fm = self.folding_coreml(patches=patches, output_size=output_size)else:fm = self.folding_pytorch(patches=patches, output_size=output_size)# MobileViTv3: local+global instead of only globalfm = self.conv_proj(torch.cat((fm, fm_conv), dim=1))# MobileViTv3: skip connectionfm = fm + xreturn fmif __name__ == '__main__':from thop import profile ## 导入thop模块model = MbViTV3(320, 160, enable_coreml_compatible_fn=False)input = torch.randn(1, 320, 44, 84)#flops, params = profile(model, inputs=(input,))outpus = model.forward_spatial(input)print('flops') ## 打印计算量# print('params', params) ## 打印参数量（2）在ultralytics/nn/tasks.py的函数parse_model中修改
def parse_model(d, ch, verbose=True): # model_dict, input_channels(3)# Parse a YOLO model.yaml dictionaryif verbose:LOGGER.info(f"\n{'':>3}{'from':>20}{'n':>3}{'params':>10} {'module':<45}{'arguments':<30}")nc, gd, gw, act = d['nc'], d['depth_multiple'], d['width_multiple'], d.get('activation')if act:Conv.default_act = eval(act) # redefine default activation, i.e. Conv.default_act = nn.SiLU()if verbose:LOGGER.info(f"{colorstr('activation:')} {act}") # printlayers, save, c2 = [], [], ch[-1] # layers, savelist, ch outfor i, (f, n, m, args) in enumerate(d['backbone'] + d['head']): # from, number, module, args.......elif m in {MbViTV3}:c2 = args[0].......
（3）在yaml配置文件中写入
# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv8 object detection model with P3-P5 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect# Parameters
nc: 2 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'# [depth, width, max_channels]n: [0.33, 0.25, 1024] # YOLOv8n summary: 225 layers, 3157200 parameters, 3157184 gradients, 8.9 GFLOPss: [0.33, 0.50, 1024] # YOLOv8s summary: 225 layers, 11166560 parameters, 11166544 gradients, 28.8 GFLOPsm: [0.67, 0.75, 768] # YOLOv8m summary: 295 layers, 25902640 parameters, 25902624 gradients, 79.3 GFLOPsl: [1.00, 1.00, 512] # YOLOv8l summary: 365 layers, 43691520 parameters, 43691504 gradients, 165.7 GFLOPsx: [1.00, 1.25, 512] # YOLOv8x summary: 365 layers, 68229648 parameters, 68229632 gradients, 258.5 GFLOPs# YOLOv8.0n backbone
backbone:# [from, repeats, module, args]- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2 320*320*64- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4 160*160*128- [-1, 3, C2f, [128, True]]- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8 80*80*256- [-1, 6, C2f, [256, True]]- [-1, 1, Conv, [512, 3, 2]] # 5-P4/16 40*40*512- [-1, 6, C2f, [512, True]]- [-1, 1, Conv, [1024, 3, 2]] # 7-P5/32 20*20*1024- [-1, 3, C2f, [1024, True]]- [-1, 1, SPPF, [1024, 5]] # 9 20*20*1024# YOLOv8.0n head
head:- [-1, 1, nn.Upsample, [None, 2, 'nearest']] # 10- [[-1, 6], 1, Concat, [1]] # 11- [-1, 3, C2f, [512]] # 12 40*40*512- [-1, 1, nn.Upsample, [None, 2, 'nearest']] # 13- [[-1, 4], 1, Concat, [1]] # 14- [-1, 3, C2f, [256]] # 15 (P3/8-small) 44*84*320- [-1, 1, MbViTV3, [320, 160]] # 16- [-1, 1, Conv, [256, 3, 2]] # 17- [[-1, 12], 1, Concat, [1]] # 18- [-1, 3, C2f, [512]] # 19 (P4/16-medium) 40*40*512- [-1, 1, Conv, [512, 3, 2]] # 20- [[-1, 9], 1, Concat, [1]] # 21- [-1, 3, C2f, [1024]] # 22 (P5/32-large) 20*20*1024- [[16, 19, 22], 1, Detect, [nc]] # 23
（4）开始训练，先把其他梯度关闭，保留新加的模块的梯度。
import os
from ultralytics import YOLO
import subprocess
from ultralytics.nn.vit.Vit import MbViTV3
os.environ['KMP_DUPLICATE_LIB_OK'] = 'True'def add_vit(model):for name, param in model.model.named_parameters():stand = name[6:8]vit_ls = ['16']if stand in vit_ls:param.requires_grad = Trueelse:param.requires_grad = Falsefor name, param in model.model.named_parameters():if param.requires_grad:print(name)return modeldef main():# model = YOLO(r'ultralytics/cfg/models/v8/yolov8x.yaml').load('/root/autodl-tmp/yolov8x.pt')model = YOLO(r'yolov8x_vit.yaml').load('runs/detect/vit/weights/vit.pt')model = add_vit(model)model.train(data="data.yaml", imgsz=640, epochs=50, batch=10, device=0, workers=0)
if __name__ == '__main__':main()————————————over————————————