YOLOv5 (6.0/6.1) brief summary
# Content + [1. Model Structure](#1) + [2. Data Augmentation](#2) + [3. Training Strategies](#3) + [4. Others](#4) - [4.1 Compute Losses](#41) - [4.2 Balance Losses](#42) - [4.2 Eliminate Grid Sensitivity](#43) - [4.3 Build Targets](#44) <a name="1"></a> ## 1. Model Structure YOLOv5 (v6.0/6.1) consists of: - **Backbone**: `New CSP-Darknet53` - **Neck**: `SPPF`, `New CSP-PAN` - **Head**: `YOLOv3 Head` Model structure (`yolov5l.yaml`):  Some minor changes compared to previous versions: 1. Replace the `Focus` structure with `6x6 Conv2d`(more efficient, refer #4825) 2. Replace the `SPP` structure with `SPPF`(more than double the speed) <details> <summary>test code</summary> ```python import time import torch import torch.nn as nn class SPP(nn.Module): def __init__(self): super().__init__() self.maxpool1 = nn.MaxPool2d(5, 1, padding=2) self.maxpool2 = nn.MaxPool2d(9, 1, padding=4) self.maxpool3 = nn.MaxPool2d(13, 1, padding=6) def forward(self, x): o1 = self.maxpool1(x) o2 = self.maxpool2(x) o3 = self.maxpool3(x) return torch.cat([x, o1, o2, o3], dim=1) class SPPF(nn.Module): def __init__(self): super().__init__() self.maxpool = nn.MaxPool2d(5, 1, padding=2) def forward(self, x): o1 = self.maxpool(x) o2 = self.maxpool(o1)
Posted by WZMIAOMIAO on Mar 16, 2022, 4:39 AM
Tags
Related issues
- struct's lifetime requirements can be bypassed via covariant field assignment
- Proposal to Reorganize the Moderation Team and Review Moderation Policies
- Progressive FPS degradation appears to be related to dota2.com — blocking the domain prevents the performance loss
- proposal: container/...: generic collection types
- Better Custom Endpoint BYOK
- A concrete policy for backward compatibility until Godot 5.0
Raise real-world problems, find people facing the same issue, and discuss solutions.2,500 issues0 comments
github (Other)
struct's lifetime requirements can be bypassed via covariant field assignment
Proposal to Reorganize the Moderation Team and Review Moderation Policies
Progressive FPS degradation appears to be related to dota2.com — blocking the domain prevents the performance loss
proposal: container/...: generic collection types
Better Custom Endpoint BYOK
A concrete policy for backward compatibility until Godot 5.0
New MTP models not supported
Tracking: 429 / Capacity Issues
YOLOv5 (6.0/6.1) brief summary
# Content + [1. Model Structure](#1) + [2. Data Augmentation](#2) + [3. Training Strategies](#3) + [4. Others](#4) - [4.1 Compute Losses](#41) - [4.2 Balance Losses](#42) - [4.2 Eliminate Grid Sensitivity](#43) - [4.3 Build Targets](#44) <a name="1"></a> ## 1. Model Structure YOLOv5 (v6.0/6.1) consists of: - **Backbone**: `New CSP-Darknet53` - **Neck**: `SPPF`, `New CSP-PAN` - **Head**: `YOLOv3 Head` Model structure (`yolov5l.yaml`):  Some minor changes compared to previous versions: 1. Replace the `Focus` structure with `6x6 Conv2d`(more efficient, refer #4825) 2. Replace the `SPP` structure with `SPPF`(more than double the speed) <details> <summary>test code</summary> ```python import time import torch import torch.nn as nn class SPP(nn.Module): def __init__(self): super().__init__() self.maxpool1 = nn.MaxPool2d(5, 1, padding=2) self.maxpool2 = nn.MaxPool2d(9, 1, padding=4) self.maxpool3 = nn.MaxPool2d(13, 1, padding=6) def forward(self, x): o1 = self.maxpool1(x) o2 = self.maxpool2(x) o3 = self.maxpool3(x) return torch.cat([x, o1, o2, o3], dim=1) class SPPF(nn.Module): def __init__(self): super().__init__() self.maxpool = nn.MaxPool2d(5, 1, padding=2) def forward(self, x): o1 = self.maxpool(x) o2 = self.maxpool(o1)
Comments (0)
No comments yet. Start the discussion.