Prueba diego

[fdobad]

No description provided.

[Copilot]

Pull Request Overview

This PR adds core configuration, model definition, evaluation logic, and dataset listings for wildfire scar segmentation experiments.

• Introduces parameters.py for dataset paths and statistical parameters used in training/evaluation
• Implements a U-Net in model_u_net.py with an exported model instance
• Adds evaluation.py to compute segmentation metrics and generate diagnostic plots
• Includes CSV files listing validation and test samples

Reviewed Changes

Copilot reviewed 18 out of 18 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
parameters.py	Defines statistical arrays and file paths for training and evaluation
model_u_net.py	Adds U-Net architecture and exports a prebuilt model instance
evaluation.py	Implements evaluation pipeline with metric computation and plotting
datasets_csv_11_2023/val_val_196.csv	Provides 196 validation samples
datasets_csv_11_2023/val_test_97.csv	Provides 97 test samples

Comments suppressed due to low confidence (4)

parameters.py:48

• Avoid hard-coding absolute, OS-specific paths; consider making this configurable or relative to the project root.

photo_results_path = "C:/Users/56965/Documents/TesisIan/.../evaluation_results/"

model_u_net.py:135

• [nitpick] Instantiating a model at import time can be surprising; consider exposing the class and letting callers instantiate it in their own context.

model = UNet(n_channels=16, n_classes=1)

evaluation.py:23

• [nitpick] There are no tests covering the evaluation logic; consider adding unit tests for obtain_model_size and the main evaluation flow.

def main(argv=None):

parameters.py:1

• [nitpick] Add a module-level docstring explaining the purpose of these parameters and how they should be used.

# ---

[Copilot]

There are large blocks of commented-out code inside unassigned triple quotes; remove or move these to real comments to reduce confusion.

Suggested change

	'''
	min_as=[0.0, 0.0, 0.0, 17.0, 7.0, 0.0, -0.0961538, -597.96850,
	0.0, 0.0, 0.0, 0.0, 8.0, 0.0, -0.0966292, -392.023010]
	max_as=[1689.0, 2502.0, 3260.0, 5650.0, 5282.0, 4121.0, 1.0, 1000.0, 1750.0,
	2559.0, 3325.0, 6065.0, 5224.0, 3903.0, 1.0, 1000.0]
	'''
	# min_as=[0.0, 0.0, 0.0, 17.0, 7.0, 0.0, -0.0961538, -597.96850,
	# 0.0, 0.0, 0.0, 0.0, 8.0, 0.0, -0.0966292, -392.023010]
	# max_as=[1689.0, 2502.0, 3260.0, 5650.0, 5282.0, 4121.0, 1.0, 1000.0, 1750.0,
	# 2559.0, 3325.0, 6065.0, 5224.0, 3903.0, 1.0, 1000.0]

[Copilot]

The gc import is unused; remove it to clean up dependencies.

Suggested change

import gc

[Copilot]

Wrap inference in with torch.no_grad(): to avoid building the computation graph and reduce memory usage during evaluation.

Suggested change

	output = model(x).cpu()
	with torch.no_grad():
	output = model(x).cpu()

[Copilot]

[nitpick] This function is very large and nested inside main; consider breaking it into smaller, top-level functions for clarity and testability.

Suggested change

	def main(argv=None):
	def obtain_model_size(input_str):
	# Define el patrón de búsqueda para '128' y 'as' en las posiciones específicas
	patron_128 = re.compile(r'_\d+_(\d+)_')
	patron_as = re.compile(r'_(as)_')
	# Busca el patrón en la cadena
	coincidencia_128 = patron_128.search(input_str)
	coincidencia_as = patron_as.search(input_str)
	# Asigna los valores a las variables según las coincidencias
	valor_128 = coincidencia_128.group(1) if coincidencia_128 else None
	valor_as = coincidencia_as.group(1) if coincidencia_as else None
	return valor_128, valor_as
	if argv is None:
	argv = sys.argv[1:]
	args = get_evaluation_args(argv)
	evald1=evald2=dataset=pd.DataFrame()
	print(f'ev1: {args.ev1}, ev2: {args.ev2}, mp: {args.mp}')
	evald1=pd.read_csv(args.ev1)
	evald2=pd.read_csv(args.ev2)
	dataset=pd.concat([evald1,evald2],axis=0,ignore_index=True)
	model_size = args.ms.upper()
	if not model_size:
	if obtain_model_size(args.mp)[0] == "128":
	model_size = "128"
	elif obtain_model_size(args.mp)[1] == "as":
	model_size = "AS"
	device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
	# Loads a model of a specific epoch to evaluate
	model.load_state_dict(torch.load(args.mp, map_location=torch.device('cpu')))
	def evaluation(model_size):
	"""
	Evaluates the metrics of the given dataset and plots images for each input comparing the pre and post-fire images and
	the original firescar vs the model's prediction.
	dataset (object): Pandas dataframe with the data's filenames from two different regions. There are 3 columns with the required data filenames for
	each input. "ImPosF": The image post Fire, "ImgPreF": The image pre Fire, and "FireScar_tif": The label, in a raster file
	model_size (str): "AS" or "128", set depending on the desired dataset, AS or 128.
	"""
	# Adjust these following parameters in the parameter's file:
	# ssx_ey: where x= index' number and y: either a for the AS model or 1 for the 128 model
	# ss1_ey (int): index of the first input from the Dataset 1: Region of Valparaiso
	# ss2_ey (int): index of the last input from the Dataset 1: Region of Valparaiso
	def evaluation(dataset, model_size, device, model):
	"""
	Evaluates the metrics of the given dataset and plots images for each input comparing the pre and post-fire images and
	the original firescar vs the model's prediction.
	dataset (object): Pandas dataframe with the data's filenames from two different regions. There are 3 columns with the required data filenames for
	each input. "ImPosF": The image post Fire, "ImgPreF": The image pre Fire, and "FireScar_tif": The label, in a raster file
	model_size (str): "AS" or "128", set depending on the desired dataset, AS or 128.
	device (torch.device): The device to run the model on.
	model (torch.nn.Module): The trained model to evaluate.
	"""
	# Adjust these following parameters in the parameter's file:
	# ssx_ey: where x= index' number and y: either a for the AS model or 1 for the 128 model
	# ss1_ey (int): index of the first input from the Dataset 1: Region of Valparaiso
	# ss2_ey (int): index of the last input from the Dataset 1: Region of Valparaiso

[Copilot]

The code mixes i and a separate cont counter for indexing test_df; unify on a single index variable to avoid confusion.

Suggested change

	test_df.loc[i,"iou"]=this_iou
	test_df.loc[cont,"iou"]=this_iou