Holds the configuration for anything you want it to. To get the currently active config, call get_cfg(). To use, just do cfg.x instead o

Keeps an average window of the specified number of items.

Composes several augmentations together. Args: transforms (List[Transform]): list of transforms to compose. Example: >>> augme

`MS Coco Detection <http://mscoco.org/dataset/#detections-challenge2016>`_ Dataset. Args: root (string): Root directory where images are d

██╗ ██╗ ██████╗ ██╗ █████╗ ██████╗████████╗ ╚██╗ ██╔╝██╔═══██╗██║ ██╔══██╗██╔════╝╚══██╔══╝ ╚████╔╝ ██║ ██║██║ █████

Stores all the information necessary to calculate the AP for one IoU and one class. Note: I type annotated this because why not.

Transforms a BRG image made of floats in the range [0, 255] to whatever input the current backbone network needs. transform is a transfo

Transorm to be used when evaluating.

Transforms a COCO annotation into a Tensor of bbox coords and label index Initilized with a dictionary lookup of classnames to indexes

Transform that does all operations on the GPU for super speed. This doesn't suppport a lot of config settings and should only be used for pro

A class that allows you to navigate a dictonary using x.a.b[2].c, etc.

A simple progress bar that just outputs a string.

If preserve_aspect_ratio is true, this resizes to an approximate area of max_size * max_size

Why is this a class? Why do I have a class for creating and parsing save paths? What am I doing with my life?

Adapted from torchvision.models.resnet

A Custom Data Parallel class that properly gathers lists of dictionaries.

This is a custom version of DataParallel that works better with our training data. It should also be faster than the general case.

An implementation of YOLOv3's Darnet53 in https://pjreddie.com/media/files/papers/YOLOv3.pdf This is based off of the implementation of

At test time, Detect is the final layer of SSD. Decode location preds, apply non-maximum suppression to location predictions based on conf sc

Implements a general version of the FPN introduced in https://arxiv.org/pdf/1612.03144.pdf Parameters (in cfg.fpn): - num_featur

This is a module version of F.interpolate (rip nn.Upsampling). Any arguments you give it just get passed along for the ride.

A class to log information during training per information and save it out. It also can include extra debug information like GPU usage / temp

SSD Weighted Loss Function Compute Targets: 1) Produce Confidence Target Indices by matching ground truth boxes with (default)

A wrapper for running the network and computing the loss This is so we can more efficiently use DataParallel.

Pads the image to the input width and height, filling the background with mean and putting the image in the top-left. Note: this expects

The (c) prediction module adapted from DSSD: https://arxiv.org/pdf/1701.06659.pdf Note that this is slightly different to the module in

Prepares the gt masks for use_gt_bboxes by cropping with the gt box and downsampling the resulting mask to mask_size, mask_size. This fun

Crop Arguments: img (Image): the image being input during training boxes (Tensor): the original bounding boxes in pt form

Transform to be used when training.

Note: channels is the lesser of the two. The output will be expansion * channels.

Applies a lambda as a transform.

Adapted from torchvision.models.resnet

Transforms a tensorized image by swapping the channels in the order specified in the swap tuple. Args: swaps (int triple): final orde

Args: - cfg: A list of layers given as lists. Layers can be either 'M' signifying a max pooling layer, a number signifyin

A class that lets you go: with timer.env(fn_name): # (...) That automatically manages a timer start and stop for you.