How liyongqingupc/mcan-oilspilldetection works
MCAN is a PyTorch oil-spill detection project trained from small image datasets. Its documented user workflow is to place training images in Input/TrainingSet and run main_train.py, then place evaluation images in Input/TestSet and run test.py with a trained model. The repository also includes multiscale image-resizing utilities, adversarial network definitions, image-generation helpers, and an Inception-based SIFID metric implementation. Training internals and entry-point wiring were not sampled, so those relationships are left unspecified.
Main entry points: main_train.py Training entry point · test.py Testing entry point · TrainingSet Training images · TestSet Test images
Components 14
Train and test
- Training images
Input/TrainingSet - Training entry point
main_train.py - Multiscale training
MCAN/training.py - Test images
Input/TestSet - Testing entry point
test.py - Run configuration
config.py
Adversarial model
- Generator and discriminator
MCAN/models.py
Image processing
- Image and GAN utilities
MCAN/functions.py - Multiscale resizing
MCAN/imresize.py - Image generation
MCAN/manipulate.py
Evaluation metric
- SIFID scoring
SIFID/sifid_score.py - Inception features
SIFID/inception.py
Other components
- Training user
- Testing user
Connections 6
- Training user → Training entry point: runs
README.md - Training images → Training entry point: provides images
README.md - Testing user → Testing entry point: runs
README.md - Test images → Testing entry point: provides images
README.md - Image and GAN utilities → Multiscale resizing: calls resizing
MCAN/functions.py - Image generation → Multiscale training: imports training
MCAN/manipulate.py