Upload folder using huggingface_hub

MODEL_FILES.md

@@ -0,0 +1,43 @@
+# MambaRefine-CD Hugging Face Model Files
+
+This manifest records the trained checkpoint files selected for the Hugging Face model repository `<HF_USERNAME_OR_ORG>/MambaRefine0cd`.
+
+## Selected Checkpoints
+
+| Dataset | Original checkpoint path | Hugging Face checkpoint path | Config file used | Metric source | SHA256 | File size | Notes |
+|---|---|---|---|---|---|---:|---|
+| WHU-CD | `outputs/whu/full/run_whu_whu_full_seed42_20260503_031818/checkpoints/best.pth` | `checkpoints/mambarefine_cd_whu_cd_best.pth` | `configs/whu_cd_run_config.yaml` | `outputs/whu/full/run_whu_whu_full_seed42_20260503_031818/test_results/test_metrics.json` | `cd3b176a483d5311c02251d2a72ac399de4aecec89ad2808df02805c89d33758` | 986.65 MiB | Best validation checkpoint, iteration 45000, EMA found, test F1 95.5324. |
+| DSIFN-CD | `archive_legacy/committed_outputs/outputs/dsifn/a6_full/run_dsifn_a6_full_seed42_20260430_105918/checkpoints/best.pth` | `checkpoints/mambarefine_cd_dsifn_cd_best.pth` | `configs/dsifn_cd_run_config.yaml` | `archive_legacy/committed_outputs/outputs/dsifn/a6_full/run_dsifn_a6_full_seed42_20260430_105918/test_results/test_metrics.json` | `6b56becbba10ad6e67e772db939339f9040de61c73dc049e3c752d04eb0bdf6f` | 987.97 MiB | Best validation checkpoint, iteration 50000, EMA found, test F1 96.3963. |
+
+## Copied Configs
+
+| Release path | Source path | Purpose |
+|---|---|---|
+| `configs/whu_cd_run_config.yaml` | `outputs/whu/full/run_whu_whu_full_seed42_20260503_031818/config.yaml` | Frozen WHU-CD run config. |
+| `configs/dsifn_cd_run_config.yaml` | `archive_legacy/committed_outputs/outputs/dsifn/a6_full/run_dsifn_a6_full_seed42_20260430_105918/config.yaml` | Frozen DSIFN-CD run config. |
+| `configs/datasets/whu.yaml` | `configs/datasets/whu.yaml` | Dataset config. |
+| `configs/datasets/dsifn.yaml` | `configs/datasets/dsifn.yaml` | Dataset config. |
+| `configs/active.yaml` | `configs/active.yaml` | Current repo config entry point. |
+| `configs/encoders/mambavision_small.yaml` | `configs/encoders/mambavision_small.yaml` | Encoder config. |
+
+## Split Evidence
+
+### WHU-CD
+
+- Dataset config name: `WHU-CD`
+- Train samples: 6096
+- Validation samples: 762
+- Test samples: 762
+
+### DSIFN-CD
+
+- Dataset config name: `DSIFN-CD`
+- Train samples: 3153
+- Validation samples: 3152
+- Test samples: Not specified in selected run manifest
+
+## TODOs
+
+- Replace `<HF_USERNAME_OR_ORG>` with the actual Hugging Face user or organization at upload time.
+- Add the official MERCon proceedings citation after publication details are available.
+- Add DSIFN-CD test split count only if it is verified from repo artifacts.

README.md

@@ -1,3 +1,130 @@
 ---
-license: mit
+license: other
+library_name: pytorch
+pipeline_tag: image-segmentation
+tags:
+- remote-sensing
+- change-detection
+- binary-change-detection
+- semantic-segmentation
+- mamba
+- mambarefine-cd
+- mercon
+metrics:
+- f1
+- iou
+- precision
+- recall
 ---
+
+# MambaRefine-CD
+
+MambaRefine-CD is a remote sensing binary change detection model. It takes a bi-temporal image pair as input and outputs a binary change mask. The paper has been accepted at MERCon. This Hugging Face repository contains trained weights, configs, and usage instructions.
+
+## Model Details
+
+- Model name: MambaRefine-CD
+- Hugging Face repo: `<HF_USERNAME_OR_ORG>/MambaRefine0cd`
+- Task: binary remote sensing change detection
+- Input: pre-change and post-change image pair
+- Output: binary change mask
+- Framework: PyTorch
+- Paper status: Accepted at MERCon
+- GitHub repository: `https://github.com/Dineth14/MambaRefine-CD`
+
+## Released Checkpoints
+
+| Dataset | Checkpoint | Config | Train Split | Validation Split | Test Split | Main Metric | Notes |
+| ------- | ---------- | ------ | ----------- | ---------------- | ---------- | ----------- | ----- |
+| WHU-CD | `checkpoints/mambarefine_cd_whu_cd_best.pth` | `configs/whu_cd_run_config.yaml` | 6096 samples | 762 samples | 762 samples | Test F1 95.5324 | Best validation checkpoint, iteration 45000, threshold 0.55, EMA found. |
+| DSIFN-CD | `checkpoints/mambarefine_cd_dsifn_cd_best.pth` | `configs/dsifn_cd_run_config.yaml` | 3153 samples | 3152 samples | Not specified in selected run manifest | Test F1 96.3963 | Best validation checkpoint, iteration 50000, threshold 0.60, EMA found. |
+
+## Datasets and Splits
+
+### WHU-CD
+
+- Official/common dataset name: WHU-CD
+- Dataset name used in config: `WHU-CD`
+- Number of image pairs: train 6096, validation 762, test 762
+- Image size: 256
+- Mask format: binary change mask
+- Binary threshold: `127` in `configs/active.yaml`
+- Ignore index: Not specified in selected WHU-CD run config
+- Normalization: ImageNet mean `[0.485, 0.456, 0.406]` and std `[0.229, 0.224, 0.225]` in `src/datasets/transforms.py`
+
+### DSIFN-CD
+
+- Official/common dataset name: DSIFN-CD
+- Dataset name used in config: `DSIFN-CD`
+- Number of image pairs: train 3153, validation 3152, test not specified in selected run manifest
+- Image size: 256
+- Mask format: binary change mask
+- Binary threshold: `127` in `configs/active.yaml`
+- Ignore index: Not specified in selected DSIFN-CD binary release run config
+- Normalization: ImageNet mean `[0.485, 0.456, 0.406]` and std `[0.229, 0.224, 0.225]` in `src/datasets/transforms.py`
+
+## Results
+
+| Dataset | Precision | Recall | F1 | IoU | OA | Notes |
+| ------- | --------: | -----: | -: | --: | -: | ----- |
+| WHU-CD | 96.0072 | 95.0623 | 95.5324 | 91.4469 | 99.5715 | From selected WHU-CD test metrics. |
+| DSIFN-CD | 96.2591 | 96.5340 | 96.3963 | 93.0434 | 97.4721 | From selected DSIFN-CD test metrics. |
+
+## Installation
+
+```bash
+git clone https://github.com/Dineth14/MambaRefine-CD
+cd MambaRefine-CD
+pip install -r requirements.txt
+pip install huggingface_hub
+```
+
+## Download Weights
+
+```python
+from huggingface_hub import hf_hub_download
+
+ckpt_path = hf_hub_download(
+    repo_id="<HF_USERNAME_OR_ORG>/MambaRefine0cd",
+    filename="checkpoints/mambarefine_cd_whu_cd_best.pth",
+)
+print(ckpt_path)
+```
+
+## Loading the Model
+
+```python
+from src.engine.checkpoint import load_checkpoint
+from src.models.build import build_model
+from src.utils.config import load_config
+from src.utils.device import get_device
+
+cfg = load_config("configs/active.yaml")
+device = get_device(cfg)
+model = build_model(cfg).to(device)
+checkpoint = load_checkpoint("checkpoints/mambarefine_cd_whu_cd_best.pth", model)
+model.eval()
+```
+
+## Evaluation
+
+The repository scripts read `configs/active.yaml`.
+
+```bash
+python val.py
+python test.py
+python infer.py
+```
+
+Before running evaluation with downloaded weights, set `checkpoint.path` in `configs/active.yaml` to the downloaded checkpoint path and set the dataset root to your local dataset.
+
+## Limitations
+
+- The weights are intended for remote sensing binary change detection.
+- Results depend on dataset domain, resolution, preprocessing, and split consistency.
+- Users should evaluate with the same preprocessing and splits used during training.
+- DSIFN-CD test split count for the selected release run was not specified in the selected run manifest.
+
+## Citation
+
+Official citation will be added after the MERCon proceedings information is available.

checkpoints/mambarefine_cd_dsifn_cd_best.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b56becbba10ad6e67e772db939339f9040de61c73dc049e3c752d04eb0bdf6f
+size 1035963870

checkpoints/mambarefine_cd_whu_cd_best.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cd3b176a483d5311c02251d2a72ac399de4aecec89ad2808df02805c89d33758
+size 1034581134

configs/active.yaml

@@ -0,0 +1,85 @@
+project:
+  name: MambaRefine-CD
+  output_root: outputs
+  seed: 42
+
+data:
+  dataset_name: DSIFN-CD
+  root: datasets/DSIFN-CD
+  image_size: 256
+  train_dir: train
+  val_dir: val
+  test_dir: test
+  a_folder: A
+  b_folder: B
+  mask_folder: Mask
+  binary_threshold: 127
+  check_split_overlap: true
+  check_hash_overlap: true
+
+model:
+  name: MambaRefineCD
+  encoder_family: mambavision
+  encoder_variant: small
+  encoder_pretrained: true
+  freeze_encoder: false
+  unfreeze_after_iters: null
+  decoder_channels: 128
+
+ablation:
+  id: original
+  temporal_input_mode: abs_signed   # abs_only | signed_only | abs_signed
+
+train:
+  device: cuda:0
+  iterations: 50000
+  batch_size: 8
+  num_workers: 8
+  lr: 0.0001
+  weight_decay: 0.01
+  optimizer: adamw
+  scheduler: cosine
+  warmup_iters: 1000
+  amp: true
+  grad_clip_norm: 1.0
+  log_interval: 50
+  val_interval: 5000
+  save_best_only: true
+  best_metric: F1
+  higher_is_better: true
+
+loss:
+  bce_weight: 1.0
+  dice_weight: 1.0
+  aux_weight: 0.2
+  boundary_weight: 0.2
+  residual_reg_weight: 0.02
+
+eval:
+  threshold: 0.5
+  sweep_thresholds_on_val: true
+  use_val_threshold_for_test: true
+  also_report_test_sweep: true
+  threshold_min: 0.05
+  threshold_max: 0.95
+  threshold_step: 0.05
+  save_predictions: true
+  measure_fps: true
+
+checkpoint:
+  path: null
+  auto_find_latest_best: true
+
+resume:
+  enabled: false
+  path: null
+  resume_optimizer: true
+  resume_scheduler: true
+  resume_iteration: true
+
+logging:
+  tensorboard: true
+  print_params: true
+  print_flops: true
+  print_peak_memory: true
+  print_fps: true

configs/datasets/dsifn.yaml

@@ -0,0 +1,6 @@
+data:
+  dataset_name: DSIFN-CD
+  root: datasets/DSIFN-CD
+  a_folder: A
+  b_folder: B
+  mask_folder: Mask

configs/datasets/whu.yaml

@@ -0,0 +1,6 @@
+data:
+  dataset_name: WHU-CD
+  root: datasets/WHU-CD
+  a_folder: A
+  b_folder: B
+  mask_folder: Mask

configs/dsifn_cd_run_config.yaml

@@ -0,0 +1,296 @@
+experiment:
+  name: a6_full
+  output_root: outputs/dsifn/a6_full
+  seed: 42
+hardware:
+  device: cuda:1
+  gpu_ids:
+  - 1
+  mixed_precision: true
+efficiency:
+  amp: true
+  amp_dtype: fp16
+  channels_last: false
+  gradient_checkpointing: false
+  compile: false
+  compile_mode: reduce-overhead
+  fast_mode: false
+  channels_multiplier: 1.0
+profiling:
+  enabled: false
+  warmup_iters: 20
+  profile_iters: 100
+  log_interval: 10
+dataloader:
+  num_workers: 8
+  pin_memory: true
+  persistent_workers: true
+  prefetch_factor: 4
+  drop_last: true
+logging:
+  train_metrics_every_iter: false
+  save_visualizations: false
+  log_interval: 20
+optimizer:
+  grad_clip_norm: 0.5
+dataset:
+  name: DSIFN-CD
+  root: /storage2/ChangeDetection/MV/Datasets/DSIFN-CD/DSIFN
+  mode: binary
+  task_type: binary_change
+  image_size: 256
+  num_workers: 8
+  pin_memory: true
+  persistent_workers: true
+  prefetch_factor: 4
+  val_ratio: 0.2
+  augmentation: true
+  batch_size: 8
+  augmentation_ops:
+  - horizontal_flip
+  - vertical_flip
+  cache_images_in_ram: false
+  cache_masks_in_ram: false
+model:
+  mode: dual
+  backbone: mambavision
+  variant: small
+  decoder: adaptive_rf
+  pretrained: true
+  output_mode: binary
+  num_classes: 1
+  freeze_backbone: false
+  drbi:
+    enabled: true
+  signed_diff:
+    enabled: true
+  cram_lite:
+    enabled: true
+    alpha: 0.5
+    apply_stages:
+    - 0
+    - 1
+    - 2
+    attention_type: spatial
+  arf_fpn:
+    enabled: true
+  boundary_refine:
+    enabled: true
+  enable_semantic_heads: false
+difference:
+  enabled: true
+  type: region_boundary
+  out_channels: 256
+  use_depthwise: true
+  gate_hidden_ratio: 0.25
+  pre_norm: true
+  use_absdiff: true
+  use_signed_diff: true
+  use_product: false
+  product_scale: 0.25
+  use_region_gate: true
+  use_boundary_gate: true
+  region_gate_min: 0.2
+  region_gate_max: 0.8
+  boundary_gate_min: 0.0
+  boundary_gate_max: 0.4
+  return_debug: false
+decoder:
+  type: adaptive_rf
+  channels: 256
+  dilation_rates:
+  - 1
+  - 2
+  - 4
+  - 8
+  use_boundary_residual: true
+  residual_scale: 0.1
+  use_depthwise: true
+  aux_weight: 0.4
+training:
+  max_iterations: 50000
+  batch_size: 8
+  lr: 5.0e-05
+  optimizer: AdamW
+  scheduler: cosine
+  warmup_iterations: 2500
+  validate_every: 5000
+  log_every: 20
+  gradient_clip: 0.5
+  weight_decay: 0.01
+  skip_nan_steps: true
+  nan_diag_every: 50
+  non_blocking_transfer: true
+  gradient_checkpointing: false
+  use_ema: true
+  ema_decay: 0.999
+  overwrite_output_dir: false
+  allow_resume_for_ablation: false
+validation:
+  batch_size: 4
+  save_samples: true
+  sample_count: 16
+  split: val
+evaluation:
+  split: test
+  threshold: 0.5
+  threshold_sweep:
+    enabled: true
+    values:
+    - 0.3
+    - 0.35
+    - 0.4
+    - 0.45
+    - 0.5
+    - 0.55
+    - 0.6
+  threshold_select_metric: F1_1
+  use_ema: true
+  inference_mode: patch
+  crop_size: 256
+  overlap: 0.25
+  use_tta: false
+  save_predictions: false
+  save_visualizations: false
+  memory_efficient: true
+  threshold_list:
+  - 0.3
+  - 0.35
+  - 0.4
+  - 0.45
+  - 0.5
+  - 0.55
+  - 0.6
+  tta_augmentations:
+  - original
+  - hflip
+  - vflip
+  - rot90
+  use_cached_predictions: false
+loss:
+  type: bce_dice
+  final:
+    type: bce_dice
+    bce_weight: 1.0
+    dice_weight: 1.0
+  coarse:
+    enabled: true
+    weight: 0.4
+  boundary:
+    enabled: true
+    weight: 0.1
+    type: bce_dice
+    target_type: sobel
+    edge_width: 3
+  bce_weight: 1.0
+  dice_weight: 1.0
+  focal_weight: 0.0
+  boundary_weight: 0.1
+  focal_gamma: 1.5
+metrics:
+  allowed:
+  - Pre
+  - Rec
+  - F1
+  - IoU
+  - OA
+  average_mode: global
+ema:
+  enabled: true
+  decay: 0.999
+  store_on_cpu: true
+resume:
+  enabled: false
+  checkpoint_path: null
+  strict: true
+checkpoint:
+  path: null
+  monitor: f1
+  mode: max
+  save_best_only: true
+  selection_metric: F1
+  save_best: true
+  save_last: true
+  save_every: null
+  save_latest: true
+  latest_every: 5000
+boundary_metrics:
+  enabled: true
+  boundary_width: 3
+  tolerance: 2
+debug:
+  enabled: false
+  output_root: debug
+  ablation_trace: false
+  name: memory_debug
+  steps: 3
+  batch_size: 8
+  image_size: 256
+  use_amp: true
+  profile_torch_ops: true
+  profile_one_step_only: true
+  compare_modes:
+  - baseline_forward_only
+  - train_forward_loss_backward
+  - train_with_return_features_false
+  - train_with_return_features_true
+  - train_with_ema_disabled
+  - train_with_ema_enabled
+  - train_with_tta_disabled
+  - eval_no_grad
+  - eval_with_tta
+  - decoder_baseline
+  - decoder_refinement
+  - decoder_adaptive_rf
+  save_memory_summary: true
+  seed: 42
+task: binary_cd
+eval:
+  split: test
+  threshold: 0.5
+  threshold_sweep:
+    enabled: true
+    values:
+    - 0.3
+    - 0.35
+    - 0.4
+    - 0.45
+    - 0.5
+    - 0.55
+    - 0.6
+  threshold_select_metric: F1_1
+  use_ema: true
+  inference_mode: patch
+  crop_size: 256
+  overlap: 0.25
+  use_tta: false
+  save_predictions: false
+  save_visualizations: false
+  memory_efficient: true
+  threshold_list:
+  - 0.3
+  - 0.35
+  - 0.4
+  - 0.45
+  - 0.5
+  - 0.55
+  - 0.6
+  tta_augmentations:
+  - original
+  - hflip
+  - vflip
+  - rot90
+  use_cached_predictions: false
+post_training: {}
+benchmark:
+  model_name: a6_full
+  output_dir: outputs/benchmark_runs/summary
+  datasets:
+  - DSIFN-CD
+  - WHU-CD
+  eval_split: test
+  main_dataset: DSIFN-CD
+  checkpoints: {}
+_meta:
+  config_path: /storage2/ChangeDetection/MV/MambaRefine-CD/configs/ablations/dsifn/a6_full.yaml
+  config_fingerprint: a38e4e9372c18e99

configs/encoders/mambavision_small.yaml

@@ -0,0 +1,4 @@
+model:
+  encoder_family: mambavision
+  encoder_variant: small
+  encoder_pretrained: true

configs/whu_cd_run_config.yaml

@@ -0,0 +1,299 @@
+experiment:
+  name: whu_full
+  output_root: outputs/whu/full
+  seed: 42
+hardware:
+  device: cuda:0
+  gpu_ids:
+  - 0
+  mixed_precision: true
+efficiency:
+  amp: true
+  amp_dtype: fp16
+  channels_last: false
+  gradient_checkpointing: false
+  compile: false
+  compile_mode: reduce-overhead
+  fast_mode: false
+  channels_multiplier: 1.0
+profiling:
+  enabled: false
+  warmup_iters: 20
+  profile_iters: 100
+  log_interval: 10
+dataloader:
+  num_workers: 8
+  pin_memory: true
+  persistent_workers: true
+  prefetch_factor: 4
+  drop_last: true
+logging:
+  train_metrics_every_iter: false
+  save_visualizations: false
+  log_interval: 20
+optimizer:
+  grad_clip_norm: 0.5
+dataset:
+  name: WHU-CD
+  root: /storage2/ChangeDetection/MV/Datasets/WHU-CD
+  split_dir: /storage2/ChangeDetection/MV/Datasets/DSIFN-CD/DSIFN/splits
+  require_explicit_splits: true
+  mode: binary
+  task_type: binary_change
+  image_size: 256
+  num_workers: 8
+  pin_memory: true
+  persistent_workers: true
+  prefetch_factor: 4
+  val_ratio: 0.2
+  augmentation: true
+  batch_size: 8
+  cache_images_in_ram: false
+  cache_masks_in_ram: false
+model:
+  mode: dual
+  backbone: mambavision
+  variant: small
+  decoder: adaptive_rf
+  pretrained: true
+  output_mode: binary
+  num_classes: 1
+  freeze_backbone: false
+  drbi:
+    enabled: true
+  signed_diff:
+    enabled: true
+  cram_lite:
+    enabled: true
+    alpha: 0.5
+    apply_stages:
+    - 0
+    - 1
+    - 2
+    attention_type: spatial
+  arf_fpn:
+    enabled: true
+  boundary_refine:
+    enabled: false
+  enable_semantic_heads: false
+difference:
+  enabled: true
+  type: region_boundary
+  out_channels: 256
+  use_depthwise: true
+  gate_hidden_ratio: 0.25
+  pre_norm: true
+  use_absdiff: true
+  use_signed_diff: true
+  use_product: false
+  product_scale: 0.25
+  use_region_gate: true
+  use_boundary_gate: true
+  region_gate_min: 0.2
+  region_gate_max: 0.8
+  boundary_gate_min: 0.0
+  boundary_gate_max: 0.4
+  return_debug: false
+decoder:
+  type: adaptive_rf
+  channels: 256
+  dilation_rates:
+  - 1
+  - 2
+  - 4
+  - 8
+  use_boundary_residual: false
+  residual_scale: 0.0
+  use_depthwise: true
+  aux_weight: 0.4
+training:
+  max_iterations: 50000
+  batch_size: 8
+  lr: 5.0e-05
+  optimizer: AdamW
+  scheduler: cosine
+  warmup_iterations: 2500
+  validate_every: 5000
+  log_every: 20
+  gradient_clip: 0.5
+  weight_decay: 0.01
+  skip_nan_steps: true
+  nan_diag_every: 50
+  non_blocking_transfer: true
+  gradient_checkpointing: false
+  use_ema: true
+  ema_decay: 0.999
+  overwrite_output_dir: false
+  allow_resume_for_ablation: false
+validation:
+  batch_size: 4
+  save_samples: true
+  sample_count: 16
+  split: val
+evaluation:
+  split: test
+  threshold: 0.5
+  threshold_sweep:
+    enabled: true
+    values:
+    - 0.3
+    - 0.35
+    - 0.4
+    - 0.45
+    - 0.5
+    - 0.55
+    - 0.6
+  threshold_select_metric: F1_1
+  use_ema: true
+  inference_mode: patch
+  crop_size: 256
+  overlap: 0.25
+  use_tta: false
+  save_predictions: false
+  save_visualizations: false
+  memory_efficient: true
+  log_mask_debug: true
+  save_debug_outputs: false
+  threshold_list:
+  - 0.3
+  - 0.35
+  - 0.4
+  - 0.45
+  - 0.5
+  - 0.55
+  - 0.6
+  tta_augmentations:
+  - original
+  - hflip
+  - vflip
+  - rot90
+  use_cached_predictions: false
+loss:
+  type: bce_dice
+  final:
+    type: bce_dice
+    bce_weight: 1.0
+    dice_weight: 1.0
+  coarse:
+    enabled: true
+    weight: 0.4
+  boundary:
+    enabled: true
+    weight: 0.1
+    type: bce_dice
+    target_type: sobel
+    edge_width: 3
+  bce_weight: 1.0
+  dice_weight: 1.0
+  focal_weight: 0.0
+  boundary_weight: 0.1
+  focal_gamma: 1.5
+metrics:
+  allowed:
+  - Pre
+  - Rec
+  - F1
+  - IoU
+  - OA
+  average_mode: global
+ema:
+  enabled: true
+  decay: 0.999
+  store_on_cpu: true
+resume:
+  enabled: false
+  checkpoint_path: null
+  strict: true
+checkpoint:
+  path: null
+  monitor: f1
+  mode: max
+  save_best_only: true
+  selection_metric: F1
+  save_best: true
+  save_last: true
+  save_every: null
+  save_latest: true
+  latest_every: 5000
+boundary_metrics:
+  enabled: true
+  boundary_width: 3
+  tolerance: 2
+debug:
+  enabled: false
+  output_root: debug
+  ablation_trace: false
+  name: memory_debug
+  steps: 3
+  batch_size: 8
+  image_size: 256
+  use_amp: true
+  profile_torch_ops: true
+  profile_one_step_only: true
+  compare_modes:
+  - baseline_forward_only
+  - train_forward_loss_backward
+  - train_with_return_features_false
+  - train_with_return_features_true
+  - train_with_ema_disabled
+  - train_with_ema_enabled
+  - train_with_tta_disabled
+  - eval_no_grad
+  - eval_with_tta
+  - decoder_baseline
+  - decoder_refinement
+  - decoder_adaptive_rf
+  save_memory_summary: true
+  seed: 42
+task: binary_cd
+eval:
+  split: test
+  threshold: 0.5
+  threshold_sweep:
+    enabled: true
+    values:
+    - 0.3
+    - 0.35
+    - 0.4
+    - 0.45
+    - 0.5
+    - 0.55
+    - 0.6
+  threshold_select_metric: F1_1
+  use_ema: true
+  inference_mode: patch
+  crop_size: 256
+  overlap: 0.25
+  use_tta: false
+  save_predictions: false
+  save_visualizations: false
+  memory_efficient: true
+  log_mask_debug: true
+  save_debug_outputs: false
+  threshold_list:
+  - 0.3
+  - 0.35
+  - 0.4
+  - 0.45
+  - 0.5
+  - 0.55
+  - 0.6
+  tta_augmentations:
+  - original
+  - hflip
+  - vflip
+  - rot90
+  use_cached_predictions: false
+post_training: {}
+benchmark:
+  model_name: whu_full
+  output_dir: outputs/benchmark_runs/summary
+  datasets:
+  - DSIFN-CD
+  - WHU-CD
+  eval_split: test
+  main_dataset: WHU-CD
+  checkpoints: {}
+_meta:
+  config_path: /storage2/ChangeDetection/MV/MambaRefine-CD/configs/experiments/whu_full.yaml
+  config_fingerprint: aeab042b9b84225a

requirements_hf.txt

@@ -0,0 +1 @@
+huggingface_hub>=0.20.0

scripts/example_download.py

@@ -0,0 +1,12 @@
+#!/usr/bin/env python3
+"""Download a released MambaRefine-CD checkpoint."""
+
+from huggingface_hub import hf_hub_download
+
+
+REPO_ID = "<HF_USERNAME_OR_ORG>/MambaRefine0cd"
+FILENAME = "checkpoints/mambarefine_cd_whu_cd_best.pth"
+
+
+if __name__ == "__main__":
+    print(hf_hub_download(repo_id=REPO_ID, filename=FILENAME))

scripts/example_inference.py

@@ -0,0 +1,45 @@
+#!/usr/bin/env python3
+"""Minimal paired-image inference example.
+
+Run from the GitHub repository root after installing repository requirements.
+"""
+
+from __future__ import annotations
+
+import argparse
+from pathlib import Path
+
+from PIL import Image
+
+from src.engine.checkpoint import load_checkpoint
+from src.engine.inference import predict_pair
+from src.models.build import build_model
+from src.utils.config import load_config
+from src.utils.device import get_device
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--config", default="configs/active.yaml")
+    parser.add_argument("--checkpoint", required=True)
+    parser.add_argument("--image-a", required=True)
+    parser.add_argument("--image-b", required=True)
+    parser.add_argument("--output", default="prediction.png")
+    parser.add_argument("--threshold", type=float, default=None)
+    args = parser.parse_args()
+
+    cfg = load_config(args.config)
+    device = get_device(cfg)
+    model = build_model(cfg).to(device)
+    checkpoint = load_checkpoint(args.checkpoint, model)
+    threshold = args.threshold
+    if threshold is None:
+        threshold = float(checkpoint.get("best_threshold", cfg.eval.threshold))
+    model.eval()
+    pred = predict_pair(model, Path(args.image_a), Path(args.image_b), cfg, device, threshold)
+    Image.fromarray((pred.numpy() * 255).astype("uint8")).save(args.output)
+    print(f"Saved {args.output}")
+
+
+if __name__ == "__main__":
+    main()