How hlidmhkust/vccib works
This repository implements research models for task-oriented image communication with out-of-distribution (OOD) detection. The user-facing workflow is experiment entry point → dataset/data module → model training or evaluation, with model variants learning task representations and detection behavior. The model family includes semantic, temporal, JSCC, VIB, position-change, and early-exit variants; shared encoders/backbones and metric/distribution losses form an internal layer. Edges from the SOURCE INDEX indicate referenced dependencies, not necessarily runtime call order. Several subsystems are represented by their documented or indexed modules; their internal behavior was not fully sampled.
Main entry points: vae.py VAE model · open_set_data_loader.py Known/OOD data · mnist_module.py MNIST data module · vae_semantic.py Semantic VAE
Components 29
Experiment workflow
- Standard runs
main.py - Semantic runs
main_semantic.py - JSCC runs
main_semantic_JSCC.py - VIB runs
main_semantic_VIB.py - Temporal runs
main_semantic_temp.py - Position runs
main_semantic_temp_change_position.py - Early-exit runs
main_semantic_temp_earlyexit.py
Data preparation
- MNIST data module
data/mnist_module.py - Known/OOD data
data/open_set_data_loader.py - Data transforms
data/data_transforms.py - Dataset helpers
data/datasets.py - Image inversion
data/invert.py
Communication models
- VAE model
models/vae.py - Semantic VAE
models/vae_semantic.py - Semantic JSCC
models/vae_semantic_JSCC.py - Semantic VIB
models/vae_semantic_VIB.py - Temporal VAE
models/vae_semantic_temp.py - Position variant
models/vae_semantic_temp_change_position.py - Early-exit model
models/vae_semantic_temp_earlyexit.py
Model internals
- ResNet semantic encoder
models/resnet_encoders_semantic_temp.py - MNIST autoencoder
models/mnist_ae.py - Encoder/decoder blocks
models/decoders_encoders.py - DenseNet backbone
models/densenet.py - VGG backbone
models/vgg.py - Wide ResNet backbone
models/wide_resnet.py - Supervised contrastive loss
losses/losses.py - MMD loss
losses/mmd.py - Triplet loss
losses/modded_triplets.py
Other components
- Researcher
Connections 48
- Researcher → Standard runs: configures
main.py - Researcher → Semantic runs: configures
main_semantic.py - Researcher → JSCC runs: configures
main_semantic_JSCC.py - Researcher → VIB runs: configures
main_semantic_VIB.py - Researcher → Temporal runs: configures
main_semantic_temp.py - Researcher → Position runs: configures
main_semantic_temp_change_position.py - Researcher → Early-exit runs: configures
main_semantic_temp_earlyexit.py - Standard runs → MNIST data module: sets up
main.py - Standard runs → Known/OOD data: sets up
main.py - Standard runs → VAE model: trains/evaluates
main.py - Semantic runs → MNIST data module: sets up
main_semantic.py - Semantic runs → Known/OOD data: sets up
main_semantic.py - Semantic runs → Semantic VAE: trains/evaluates
main_semantic.py - JSCC runs → MNIST data module: sets up
main_semantic_JSCC.py - JSCC runs → Known/OOD data: sets up
main_semantic_JSCC.py - JSCC runs → Semantic JSCC: trains/evaluates
main_semantic_JSCC.py - VIB runs → MNIST data module: sets up
main_semantic_VIB.py - VIB runs → Known/OOD data: sets up
main_semantic_VIB.py - VIB runs → Semantic VIB: trains/evaluates
main_semantic_VIB.py - Temporal runs → MNIST data module: sets up
main_semantic_temp.py - Temporal runs → Known/OOD data: sets up
main_semantic_temp.py - Temporal runs → Temporal VAE: trains/evaluates
main_semantic_temp.py - Position runs → MNIST data module: sets up
main_semantic_temp_change_position.py - Position runs → Known/OOD data: sets up
main_semantic_temp_change_position.py - Position runs → Position variant: trains/evaluates
main_semantic_temp_change_position.py - Early-exit runs → MNIST data module: sets up
main_semantic_temp_earlyexit.py - Early-exit runs → Known/OOD data: sets up
main_semantic_temp_earlyexit.py - Early-exit runs → Early-exit model: trains/evaluates
main_semantic_temp_earlyexit.py - Known/OOD data → Data transforms: uses
data/open_set_data_loader.py - Known/OOD data → Dataset helpers: uses
data/open_set_data_loader.py - Known/OOD data → Image inversion: uses
data/open_set_data_loader.py - VAE model → MNIST autoencoder: references
models/vae.py - VAE model → Encoder/decoder blocks: references
models/vae.py - VAE model → DenseNet backbone: references
models/vae.py - VAE model → VGG backbone: references
models/vae.py - VAE model → Wide ResNet backbone: references
models/vae.py - VAE model → Supervised contrastive loss: references
models/vae.py - VAE model → MMD loss: references
models/vae.py - VAE model → Triplet loss: references
models/vae.py - Semantic VAE → ResNet semantic encoder: references
models/vae_semantic.py - Semantic VAE → Triplet loss: references
models/vae_semantic.py - Semantic VAE → Supervised contrastive loss: references
models/vae_semantic.py - Semantic VAE → MMD loss: references
models/vae_semantic.py - Semantic JSCC → ResNet semantic encoder: references
models/vae_semantic_JSCC.py - Semantic JSCC → Triplet loss: references
models/vae_semantic_JSCC.py - Semantic VIB → ResNet semantic encoder: references
models/vae_semantic_VIB.py - Semantic VIB → Triplet loss: references
models/vae_semantic_VIB.py - Temporal VAE → ResNet semantic encoder: uses
models/vae_semantic_temp.py