How syedaliasghar14/cnn-for-asi works
This is a PyTorch seismic-interpretation application: it trains a 3D CNN from SEG-Y seismic volumes and colored slice labels, then applies the trained model to slices or a full cube and can extract learned features. The principal workflow runs from seismic and label inputs through augmented training batches, model optimization and saved weights to prediction and SEG-Y output. TensorBoard visualization is an additional capability. Prediction entry-point implementations were not sampled, so their internal wiring is shown only where README documentation supports it.
Main entry points: train.py Training loop · utils.py Slice and cube inference · test.py Prediction example · data.py SEG-Y and labels
Components 18
Seismic inputs
- SEG-Y and labels
data.py - Seismic cube
- Label coordinates
- SEG-Y writer
data.py
Model training
- Augmented batches
batch.py - Training loop
train.py - 3D TextureNet
texture_net.py - Saved weights
Interpretation
- Prediction example
test.py - Multi-GPU prediction
test_parallel.py - Slice and cube inference
utils.py - Feature extraction
texture_net.py
Visualization
- TensorBoard logger
tb_logger.py
Other components
- Researcher
- Seismic SEG-Y
- Label images
- Interpretation SEG-Y
- TensorBoard
Connections 24
- Researcher → Seismic SEG-Y: provides data
- Researcher → Label images: provides labels
- Seismic SEG-Y → SEG-Y and labels: loads
- Label images → SEG-Y and labels: loads
- SEG-Y and labels → Seismic cube: returns cube
- SEG-Y and labels → Label coordinates: returns coordinates
- Seismic cube → Augmented batches: samples
- Label coordinates → Augmented batches: selects centers
- Augmented batches → Training loop: supplies batches
- Training loop → 3D TextureNet: trains
- Training loop → Slice and cube inference: calls inference
- Training loop → Saved weights: saves weights
- Researcher → Prediction example: runs prediction
- Researcher → Multi-GPU prediction: runs prediction
- Prediction example → Slice and cube inference: predicts slices/cube
- Multi-GPU prediction → Slice and cube inference: predicts in parallel
- Prediction example → Feature extraction: extracts features
- Slice and cube inference → 3D TextureNet: evaluates
- Prediction example → Saved weights: loads weights
- Multi-GPU prediction → Saved weights: loads weights
- Slice and cube inference → SEG-Y writer: produces interpretation
- SEG-Y writer → Interpretation SEG-Y: writes
- Training loop → TensorBoard logger: logs progress
- TensorBoard logger → TensorBoard: writes summaries