Skip to content

Pretrained Models

This page documents the pretrained checkpoints released under Pretrained_Model/.

The repository provides pretrained models for:

  • GAN-based baselines
  • UNet-based baselines

These checkpoints are intended for:

  • evaluation of released benchmark tasks
  • reproduction of reported pretrained results
  • comparison against new baselines under the same benchmark settings

GAN Checkpoints

GAN checkpoints are organized as:

Pretrained_Model/GAN/<task_id>/
  best_G.pth
  best_D.pth
  config.json
  eval_results.json
  epoch_history.json
  batch_history.json

File roles

best_G.pth

Best generator checkpoint.

This is the main checkpoint used for:

  • inference
  • evaluation
  • pretrained result reproduction

best_D.pth

Best discriminator checkpoint.

This is mainly provided for:

  • adversarial training continuation
  • full training-state reproduction

config.json

Stores the task configuration associated with the training run.

Typical contents may include:

  • split strategy
  • input mode
  • sparse / dense mode
  • number of samples
  • training hyperparameters

eval_results.json

Stores evaluation results for the pretrained checkpoint.

epoch_history.json

Stores epoch-level training history.

batch_history.json

Stores batch-level training history.

Available GAN tasks

The released pretrained GAN tasks include:

  • beam_dense_encoding
  • beam_dense_feature
  • random_dense_encoding
  • random_dense_feature
  • random_sparse_encoding_samples819
  • random_sparse_feature_samples819
  • scene_dense_encoding
  • scene_dense_feature

These task IDs are directly aligned with the benchmark naming defined in the Benchmark page.

For evaluation or inference, users typically only need:

  • best_G.pth
  • config.json

The folder naming directly indicates the benchmark task, for example:

Pretrained_Model/GAN/random_dense_feature/
Pretrained_Model/GAN/scene_dense_encoding/
Pretrained_Model/GAN/random_sparse_feature_samples819/

This makes it straightforward to map:

  • benchmark task
  • pretrained checkpoint
  • evaluation configuration

UNet Checkpoints

UNet checkpoints are stored under:

Pretrained_Model/Unet/
  *.pt
  *.png
  *.csv

File roles

*.pt

Serialized UNet checkpoints.

These files contain pretrained model weights for different released benchmark settings.

*.png

Visualization files for qualitative result inspection.

These may include:

  • prediction visualizations
  • comparison plots
  • example outputs

*.csv

Summary tables or recorded evaluation results.

UNet checkpoint usage

Unlike the GAN checkpoints, the released UNet checkpoints are not organized by <task_id> folder names.

Instead, they are evaluated through the released UNet evaluation pipeline, which uses a predefined checkpoint list and corresponding benchmark settings.

This means:

  • GAN checkpoints follow task-aligned folder naming
  • UNet checkpoints follow a script-aligned checkpoint naming style

  • both still correspond to the same released benchmark dimensions:

  • random / beam / scene

  • dense / sparse
  • feature / encoding

Released UNet model groups

The released UNet evaluation script covers multiple baseline groups, including:

  • random-split dense baselines
  • random-split sparse baselines
  • beam-split dense baselines
  • scene-split dense baselines
  • feature-map-based baselines
  • continuous-encoding-based baselines

Representative released checkpoint groups include:

  • Solution1_environment
  • Solution1_featuremap
  • Solution1_continuous
  • Solution2_sparse_featuremap
  • Solution2_sparse_continuous
  • Solution3_1_beam_featuremap
  • Solution3_1_beam_continuous
  • Solution3_2_scene_featuremap
  • Solution3_2_scene_continuous

These are evaluated directly by the released UNet evaluation script.

For most users, the recommended workflow is:

  1. place the UNet checkpoints under Pretrained_Model/Unet/
  2. run the released evaluation script:

python ModelEvaluation_Unet.py
3. use the generated summary tables and visualizations for inspection

In practice, users do not need to manually reconstruct the task mapping for every checkpoint as long as they use the released evaluation script.

Summary of Checkpoint Organization

Baseline Checkpoint organization Recommended usage
GAN Pretrained_Model/GAN/<task_id>/ directly map task → folder → checkpoint
UNet Pretrained_Model/Unet/*.pt with script-defined mapping use released UNet evaluation script

Suggested Starting Points

For first use, the following pretrained GAN models are recommended:

  • random_dense_feature — standard dense baseline
  • scene_dense_feature — cross-environment generalization benchmark
  • random_sparse_feature_samples819 — sparse reconstruction benchmark

These provide representative coverage of:

  • standard supervised prediction
  • generalization to unseen scenes
  • sparse observation settings

For UNet baselines, the recommended starting point is to run the released UNet evaluation script directly and inspect the predefined evaluated model groups.

Checkpoint-to-task examples

Baseline Example checkpoint / folder Benchmark meaning
GAN random_dense_feature/best_G.pth random split + dense + feature
GAN scene_dense_encoding/best_G.pth scene split + dense + encoding
UNet Solution2_sparse_featuremap sparse + feature
UNet Solution3_2_scene_continuous scene + dense + continuous