From a list of samples from the dataset, returns the batched images and targets. This should be passed to the DataLoader

SRPooler for Detection with or without FPN. Also, the requirement of passing the scales is not strictly necessary, as they can be inferre

A class for filtering AOT entities A gt entity will be filtered if it falls into one the following criterion 1. tracking id is not Heli

A class for filtering JTA dataset entities during evaluation A gt entity will be filtered (ignored) if its id is -1 (negative)

Deprecated, it is slow. It is only used for sanity check.

Combines a set of individual heads (for box prediction or masks) into a single head.

The default video iterator with decord: https://github.com/dmlc/decord There would be memory issue when the video length is too long (> 20 mi

Implement a wrapper to call tracker

Feature extraction for template and search region is different in this case

Sample track targets for SiamMOT. It samples from track proposals from RPN

Perform JPEG compression augmentation to an image

Crop a patch from the image and resize to its original size

Get a video iterator for a video that is represented with a image folder that includes a list of JPEG frames

Perform motion augmentation to an image

Split the video into small chunks (in an non-overlapping fashion) for inference

A class for filtering MOT dataset entities A gt entity will be filtered (ignored) if its visibility ratio is very low A predicted entity

From a set of classification scores, box regression and proposals, computes the post-processed boxes, and applies NMS to obtain the final

Generic Box Head class.

Main class for R-CNN. Currently supports boxes and tracks. It consists of three main parts: - backbone - rpn - heads: takes the f

A class to manage the track id distribution (initiate/kill a track)

A class that includes utility functions unique to track branch

Generate a video for visualization

Write the artifact of tracking results

DLA Basic

Res2Net/Res2NeXT DLA Bottleneck Adapted from https://github.com/gasvn/Res2Net/blob/master/dla.py

DLA/DLA-X Bottleneck