How whuyyx/magc works
MAGC is a map-assisted remote-sensing image compression system for extremely low bitrates. Its inference workflow loads a configured, checkpointed model, processes input images through the ControlLDM compression-and-reconstruction path, then saves results and reports bitrate and image-quality metrics. Training has two documented stages, backed by dataset loading and batch transforms. The graph also shows the latent-diffusion and entropy-coding layers that support the model. Some inference and model internals are unsampled; arrows are limited to relationships shown in excerpts, the source index, or README.
Main entry points: cldm.py ControlLDM · ddpm.py Latent diffusion · inference.py Inference runner · train.py Training entry
Components 20
User workflow
- Inference runner
inference.py - Quality metrics
cal_metrics/iqa.py - Training entry
train.py - Dataset loader
dataset/data_module.py - Batch transforms
dataset/batch_transform.py
MAGC model
- ControlLDM
model/cldm.py - Map adapter
model/adapters.py - Spaced sampler
model/spaced_sampler.py
Latent diffusion
- Latent diffusion
ldm/models/diffusion/ddpm.py - DDIM sampler
ldm/models/diffusion/ddim.py - Latent autoencoder
ldm/models/autoencoder.py - Diffusion networks
ldm/modules/diffusionmodules/openaimodel.py - Attention layers
ldm/modules/attention.py - Config instantiation
ldm/util.py
Compression runtime
- Hyperencoder
ldm/models/hyperencoder.py - Entropy models
compressai/entropy_models/entropy_models.py
Other components
- Image user
- Remote-sensing images
- Reconstructions and metrics
- Training data
Connections 21
- Image user → Inference runner: runs inference
- Remote-sensing images → Inference runner: supplies images
- Inference runner → ControlLDM: loads model
inference.py - Inference runner → Quality metrics: computes metrics
inference.py - Inference runner → Reconstructions and metrics: saves results
inference.py - ControlLDM → Latent diffusion: extends
model/cldm.py - ControlLDM → Map adapter: instantiates
model/cldm.py - ControlLDM → Hyperencoder: uses
model/cldm.py - ControlLDM → Spaced sampler: uses
model/cldm.py - Latent diffusion → Latent autoencoder: instantiates
ldm/models/diffusion/ddpm.py - Latent diffusion → DDIM sampler: imports
ldm/models/diffusion/ddpm.py - Latent diffusion → Hyperencoder: imports
ldm/models/diffusion/ddpm.py - Latent diffusion → Config instantiation: configures
ldm/models/diffusion/ddpm.py - Latent autoencoder → Diffusion networks: uses layers
ldm/models/autoencoder.py - Diffusion networks → Attention layers: uses attention
ldm/modules/diffusionmodules/openaimodel.py - Hyperencoder → Entropy models: uses entropy coding
ldm/models/hyperencoder.py - Image user → Training entry: starts training
- Training data → Dataset loader: provides samples
dataset/data_module.py - Dataset loader → Batch transforms: applies transforms
dataset/data_module.py - Training entry → Dataset loader: loads data
- Training entry → ControlLDM: trains model