❓ Engine test and predict image size is always 256 x 256 #3012
Description
Your Question
I am using EfficientAD model and configuring the image size to 512 x 512 using the function configure_pre_processor(). And I think it accepts it well. I found this out by monitoring the GPU memory while training. However the tests using engine.test() after training save the anomaly maps in 768 x 256 consisting of 3 images in one (Normal, Normal+Anomaly, Normal + Predictions). Same behavior is observed in the function engine.predict(). I would like to get each of these images in the form of 512 x 512. Am I missing anything ? Below is my code.
from pathlib import Path
from anomalib.data import Folder
from anomalib.engine import Engine
from anomalib.models import EfficientAd
# 2. Set up a pre-processing configuration
pre_processor = EfficientAd.configure_pre_processor(image_size=(512, 512))
# 3. Custom Configuration
# Configure model parameters
model = EfficientAd(
pre_processor=pre_processor,
)
# 4. Training Pipeline
# Default structure expects:
# - train/good: Normal (good) training images
# - test/good: Normal test images
# - test/defect: Anomalous test images
datamodule = Folder(
name="hasse",
root=Path("./datasets/my_dataset"),
normal_dir="OK", # Subfolder containing normal images
abnormal_dir="NOK", # Subfolder containing anomalous images
train_batch_size=1,
)
# Initialize training engine with specific settings
engine = Engine(
max_epochs=20,
accelerator="auto", # Automatically detect GPU/CPU
devices=1, # Number of devices to use
)
if __name__ == "__main__":
# Train the model
engine.fit(
model=model,
datamodule=datamodule,
)
# Test the model performance
test_results = engine.test(
model=model,
datamodule=datamodule,
)
Forum Check
- I have searched the discussions forum for an answer to my question.