feat: add darkness filtering capability to dataloader

[maharajamihir]

• Introduced a DarknessFilter class to filter out sequences with images that are too dark based on a specified threshold.
• Added a darkness_threshold parameter to Args and the get_dataloader function to support this feature.
• DarknessFilter is optional. A threshold value greater than 0.0 activates it.

[Copilot]

Pull Request Overview

This PR adds darkness filtering capability to the dataloader to filter out video sequences containing images that are too dark based on a configurable threshold.

• Introduced a DarknessFilter class that computes average brightness of image sequences and filters them based on a threshold
• Added darkness_threshold parameter to the Args class and get_dataloader function with a default value of 0.0
• Integrated the darkness filter into the data processing pipeline between episode processing and batching

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
utils/dataloader.py	Implements DarknessFilter class and adds darkness filtering to the dataloader pipeline
train_dynamics.py	Adds darkness_threshold parameter to configuration and passes it to the dataloader

[Copilot]

The DarknessFilter is always instantiated even when darkness_threshold is 0.0 (disabled). Consider conditionally adding the filter only when darkness_threshold > 0.0 to avoid unnecessary processing overhead.

Suggested change

	]
	]
	if darkness_threshold > 0.0:
	operations.append(
	DarknessFilter(
	darkness_threshold=darkness_threshold
	)
	)
	operations.append(
	grain.transforms.Batch(batch_size=per_process_batch_size, drop_remainder=True)
	)

[Copilot]

Using print() statements for logging in production code is not recommended. Consider using a proper logging framework like Python's logging module to allow for better control over log levels and output destinations.

Suggested change

	print(
	logging.info(

[Copilot]

The RGB to grayscale conversion coefficients [0.2989, 0.5870, 0.1140] are magic numbers. Consider defining these as named constants (e.g., RGB_TO_GRAYSCALE_WEIGHTS) to improve code readability and maintainability.

Suggested change

	element_greyscale = np.dot(element[...,:3], [0.2989, 0.5870, 0.1140])
	element_greyscale = np.dot(element[...,:3], RGB_TO_GRAYSCALE_WEIGHTS)

[emergenz]

agreed, this is a magic number