How yoon307/dig works
DiG is a research implementation of diffusion-guided weakly supervised semantic segmentation. The documented workflow trains on VOC or COCO image-level labels, produces class activation maps (CAMs) through inference, then evaluates CAM outputs. Its model combines transformer-based recognition with diffusion-guided components. The graph also preserves dataset-specific evaluation and the diffusion implementation. Relationships are limited to README statements and source-index references or excerpts; several runtime details were not sampled, so uncertain links are omitted.
Main entry points: model_ECCV24_DiG.py DiG model · train_trm_coco.py COCO training · infer_trm.py CAM inference · mctformer.py MCTformer
Components 18
Training and evaluation
- VOC training
train_trm.py - COCO training
train_trm_coco.py - CAM inference
infer_trm.py
Datasets and transforms
- VOC data
voc12/data.py - COCO data
coco14/data.py - Image transforms
tools/imutils.py - Dataset builder
tools/utils.py
Segmentation model
- DiG model
models/model_ECCV24_DiG.py - MCTformer
networks/mctformer.py - Vision transformer
networks/vision_transformer.py - ResNet-101
networks/resnet101.py - ResNet-38d
networks/resnet38d.py
Diffusion components
- Gaussian diffusion
denoising_diffusion_pytorch/denoising_diffusion_pytorch.py - FID evaluation
denoising_diffusion_pytorch/fid_evaluation.py
CAM outputs and metrics
- VOC CAM evaluation
evaluation.py - COCO CAM evaluation
evaluation_coco.py - CAM artifacts
Other components
- Researcher
Connections 22
- Researcher → VOC training: starts training
- Researcher → COCO training: starts training
- VOC training → VOC data: trains on
- COCO training → COCO data: trains on
- Dataset builder → VOC data: builds dataset
tools/utils.py - Dataset builder → COCO data: builds dataset
tools/utils.py - VOC data → Image transforms: uses transforms
voc12/data.py - COCO data → Image transforms: uses transforms
coco14/data.py - COCO training → DiG model: creates model
train_trm_coco.py - DiG model → MCTformer: uses backbone
models/model_ECCV24_DiG.py - MCTformer → Vision transformer: uses transformer
networks/mctformer.py - MCTformer → ResNet-101: uses backbone
networks/mctformer.py - Vision transformer → ResNet-38d: references
networks/vision_transformer.py - DiG model → Gaussian diffusion: uses diffusion
models/model_ECCV24_DiG.py - Gaussian diffusion → FID evaluation: evaluates FID
denoising_diffusion_pytorch/denoising_diffusion_pytorch.py - Researcher → CAM inference: runs inference
- CAM inference → DiG model: loads model
infer_trm.py - CAM inference → CAM artifacts: writes CAMs
- Researcher → VOC CAM evaluation: runs evaluation
- Researcher → COCO CAM evaluation: runs evaluation
- VOC CAM evaluation → CAM artifacts: evaluates CAMs
- COCO CAM evaluation → CAM artifacts: evaluates CAMs