Simplified Conv BN ReLU

Conv and BN with Hardswish activation

This code is based on https://github.com/facebookresearch/detr/blob/main/models/detr.py

Attention mapps queries and a set of key-value pairs to outputs, and Multi-Head Attention performs multiple parallel attention to jointly att

Generalized Intersection over Union, see https://arxiv.org/abs/1902.09630 Args: loss_weight (float): giou loss weight, default as 1

C2f layer with 2 convs, named C2f in YOLOv8

RepLayer with RepConvs, like CSPLayer(C3) in YOLOv5/YOLOX named RepBlock in YOLOv6

Standard convolution in BepC3-Block

C2 layer with 2 convs, named C2 in YOLOv8

ELAN layer used in YOLOv7, like CSPLayer(C3) in YOLOv5/YOLOX

Normal Transpose, default for upsampling

Using CIB in C2f

see https://arxiv.org/pdf/2205.12740.pdf Args: loss_weight (float): siou loss weight, default as 1 eps (float): epsilon to a

r""" LayerNorm that supports two data formats: channels_last (default) or channels_first. The ordering of the dimensions in the inputs. channels_

Args: input_shape (int): network input image size scale_factor (float): scale factor of ori image

Spatial Pyramid Pooling - Fast (SPPF) layer used in YOLOv5 by Glenn Jocher, equivalent to SPP(k=(5, 9, 13))

Single level attribute dict, NOT recursive

CSP (Cross Stage Partial) layer with 3 convs, named C3 in YOLOv5

CSPNeXt layer used in RTMDet, like CSPLayer(C3) in YOLOv5/YOLOX

Conv and BN without activation

iou loss, see https://arxiv.org/abs/1908.03851 loss = 1.0 - iou * iou Args: loss_weight (float): iou loss weight, default is 2.5

MPConvLayer used in YOLOv7

Attention layer using in YOLOv10

Build common distill model. Args: cfg: The student config. slim_cfg: The teacher and distill config.

Focus width and height information into channel space, used in YOLOX.

A LayerNorm variant, popularized by Transformers, that performs point-wise mean and variance normalization over the channel dimension for inp

Exponential Weighted Average for Deep Neutal Networks Args: model (nn.Layer): Detector of model. decay (int): The decay used

Fix the backbones variable names for pretrained weight

r"""Quality Focal Loss (QFL) is a variant of `Generalized Focal Loss: Learning Qualified and Distributed Bounding Boxes for Dense Object Detec

resize image by target_size and max_size Args: target_size (int): the target size of image keep_ratio (bool): whether keep_ratio o

Spatial-channel decoupled downsampling layer, named SCDown in YOLOv10

SPPF with BaseConv, use silu

Spatial Pyramid Pooling (SPP) layer used in YOLOv3-SPP and YOLOX

Simplified CSP SPPF with SimConv, use relu, YOLOv6 v3.0 added

Simplified SPPF with SimConv, use relu

Use VisualDL to log data or image

r""" The AdamWDL optimizer is implemented based on the AdamW Optimization with dynamic lr setting. Generally it's used for transformer model.