deepdisc.astrodet.detectron

Classes

SaveHook	This Hook saves the model during training
LossEvalHook	Validation loss code adopted from https://gist.github.com/ortegatron/c0dad15e49c2b74de8bb09a5615d9f6b
CustomLRScheduler	A hook which executes a torch builtin LR scheduler and summarizes the LR.
GenericWrapperTransform	Code from Deshwal on Stack Overflow
CustomAug	Given a probability and a custom function, return a GenericWrapperTransform object whose apply_image
ShapeAug	Given a probability and a custom function, return a GenericWrapperTransform object whose apply_image
KRandomAugmentationList	Select and Apply "K" augmentations in "RANDOM" order with "Every" __call__ method invoke
_TransformToAug

Functions

plot_stretch_Q(dataset_dicts, astro_metadata[, num, ...])	Plots different normalizations of your image using the stretch, Q parameters.
_transform_to_aug(tfm_or_aug)	Wrap Transform into Augmentation.

Module Contents

plot_stretch_Q(dataset_dicts, astro_metadata, num=0, stretches=[0.01, 0.1, 0.5, 1], Qs=[1, 10, 5, 100], ceil_percentile=99.5)

Plots different normalizations of your image using the stretch, Q parameters.

Parameters:

• dataset_dicts (dict) – detectron dataset dictionary

• num (int) – Dataset number/index to use

• stretches (array) – List of stretch params you want to permutate through to find optimal image normalization. Default is [0.01, 0.1, 0.5, 1]

• Qs (array) – List of Q params you want to permutate through to find optimal image normalization. Default is [1, 10, 5, 100]

• from (Code adapted) – https://pmelchior.github.io/scarlet/tutorials/display.html

Returns:

fig

Return type:

Figure object

class SaveHook(save_period)

Bases: detectron2.engine.HookBase

This Hook saves the model during training

Parameters:

save_period (int) – How many iterations to run before a checkpoint is saved

output_name = 'model_temp'

_period

set_output_name(name)

after_step()

class LossEvalHook(eval_period, model, data_loader)

Bases: detectron2.engine.HookBase

Validation loss code adopted from https://gist.github.com/ortegatron/c0dad15e49c2b74de8bb09a5615d9f6b

Parameters:

• eval_period (int) – How many iterations to run before validation loss is calculated

• model (torch.NN.module) – The model being trained

• data_loader (detectron2 DataLoader) – The dataloader that loads in the evaluation dataset

_model

_period

_data_loader

_do_loss_eval()

_get_loss(data)

after_step()

class CustomLRScheduler(optimizer=None, scheduler=None)

Bases: detectron2.engine.HookBase

A hook which executes a torch builtin LR scheduler and summarizes the LR. It is executed after every iteration.

Parameters:

• optimizer (torch optimizer) – the training optimizer

• scheduler (torch scheduler) – The learning rate scheduler

_optimizer = None

_scheduler = None

before_train()

static get_best_param_group_id(optimizer)

after_step()

property scheduler

state_dict()

load_state_dict(state_dict)

class GenericWrapperTransform(custom_function)

Bases: fvcore.transforms.transform.Transform

Code from Deshwal on Stack Overflow Generic wrapper for any transform (for color transform only. You can give functionality to apply_coods, apply_segmentation too)

apply_image(img)

apply transformation to image array based on the custom_function

apply_coords(coords)

Apply transformations to Bounding Box Coordinates. Currently is won’t do anything but we can change this based on our use case

inverse()

apply_segmentation(segmentation)

Apply transformations to segmentation. currently is won’t do anything but we can change this based on our use case

class CustomAug(custom_function, prob=1.0)

Bases: detectron2.data.transforms.Augmentation

Given a probability and a custom function, return a GenericWrapperTransform object whose apply_image will be called to perform augmentation

get_transform(image)

Based on probability, choose whether you want to apply the given function or not

class ShapeAug(custom_function, prob=1.0)

Bases: detectron2.data.transforms.Augmentation

Given a probability and a custom function, return a GenericWrapperTransform object whose apply_image will be called to perform augmentation

get_transform(image, instances)

Based on probability, choose whether you want to apply the given function or not

class KRandomAugmentationList(augs, k: int = -1, cropaug=None)

Bases: detectron2.data.transforms.Augmentation

Select and Apply “K” augmentations in “RANDOM” order with “Every” __call__ method invoke

max_range

k = -1

augs

cropaug = None

_setup_augs(augs, k: int)

Setup the argument list. Generates the list of argument to use from the given list :param augs: list of [Augmentation or Transform]) :param k: Number of augment to use from the given list in range [1,len_augs]. :param If False: :param use all. If it is -1: :param generate K randomly between [1: :param len_augs]:

__call__(aug_input) → fvcore.transforms.transform.Transform

__repr__()

__str__

class _TransformToAug(tfm: fvcore.transforms.transform.Transform)

Bases: detectron2.data.transforms.Augmentation

tfm

get_transform(*args)

__repr__()

__str__

_transform_to_aug(tfm_or_aug)

Wrap Transform into Augmentation. Private, used internally to implement custom augmentations.