A Conv2d -> Batchnorm -> silu/leaky relu block

Generation of cropped input person and pose heatmaps from SimplePose. Parameters ---------- img: torch.Tensor A tensor with shape

Lightweight 3x3 convolution. 1x1 (linear) + dw 3x3 (nonlinear).

C3 in yolov5, CSP Bottleneck with 3 convolutions

Residual network. Reference: - He et al. Deep Residual Learning for Image Recognition. CVPR 2016. - Xie et al. Aggregated Res

1x1 convolution + bn + relu.

Omni-Scale Network. Reference: - Zhou et al. Omni-Scale Feature Learning for Person Re-Identification. ICCV, 2019. - Zhou et

1x1 convolution + bn (w/o non-linearity).

Generation of cropped input person, pose coords, smpl parameters. Parameters ---------- img: torch.Tensor A tensor with shape: `(

YOLOv3 model. Darknet 53 is the default backbone of this model.

YOLOX model module. The module list is defined by create_yolov3_modules function. The network returns loss values from three YOLO layers duri

1x1 convolution + bn + relu.

Initialize: inplanes, planes, upscale_factor OUTPUT: (planes // upscale_factor^2) * ht * wd

Separable Conv

RetinaNet Anchors class.

A mini-network that generates channel-wise gates conditioned on input tensor.

A config utility class.

1x1 convolution + bn (w/o non-linearity).

Initialize: inplanes, planes, upscale_factor OUTPUT: (planes // upscale_factor^2) * ht * wd

YOLOv3 object detection model

Average data logger.

Placeholder for 'route' and 'shortcut' layers

A simple Kalman filter for tracking bounding boxes in image space. The 8-dimensional state space x, y, a, h, vx, vy, va, vh

A simple Kalman filter for tracking bounding boxes in image space. The 8-dimensional state space x, y, a, h, vx, vy, va, vh

Lightweight 3x3 convolution. 1x1 (linear) + dw 3x3 (nonlinear).

Lightweight convolution stream.

SEResnet

Spatial pyramid pooling layer used in YOLOv3-SPP

Labeler for multiscale anchor boxes.

A mini-network that generates channel-wise gates conditioned on input tensor.

Convolution layer (conv + bn + relu).

Convolution layer (conv + bn + relu).

Focus width and height information into channel space.

Class to store results from the matcher. This class is used to store the results from the matcher. It provides convenient methods to query th

Omni-Scale Network. Reference: - Zhou et al. Omni-Scale Feature Learning for Person Re-Identification. ICCV, 2019. - Zhou

Initialize: inplanes, planes, upscale_factor OUTPUT: (planes // upscale_factor^2) * ht * wd

Randomly selects a rectangle region in an image and erases its pixels. 'Random Erasing Data Augmentation' by Zhong et al. See https:/

Residual layer with `in_channels` inputs.

Spatial Attention Layer