Upload folder using huggingface_hub

README.md

@@ -61,12 +61,13 @@ project_root/
 │   ├── sliding_window.py        # Core logic for windowed inference
 │   └── export_logits.py         # Export of softmax probabilities
 ├── projection/
-│   └── lidar_projection.py      # Projection of predictions to LiDAR space
+│   ├── lidar_projection.py      # Projection of predictions to LiDAR space
+│   └── fast_proj.py             # Utilities for projection (Agsoft conventions), accelerated with Jax
 ├── utils/
 │   ├── logging_utils.py         # Logging setup
 │   ├── metrics.py               # Evaluation metrics (IoU, F1)
-│   ├── morton.py                # Morton code utility
 │   └── seed.py                  # Reproducibility utilities
+├── best_model.pth               # Weights for best model
 └── requirements.txt             # Python dependencies
 ```
 
@@ -163,9 +164,9 @@ The following environment was used to train and evaluate the baseline model:
 | Transformers    | 🤗 Transformers 4.51             |
 | JAX             | jax==0.6.0                       |
 | laspy           | >= 2.0                           |
-| RAM             | ≥ 64 GB recommended              |
+| RAM             | 256 GB (≥ 64 GB recommended)     |
 
-⚠️ For operations involving batch sliding-window inference and 3D projection with JAX on large scenes, high VRAM (≥ 24 GB). If CUDA OOM error, decrease:
+⚠️ For operations involving batch sliding-window inference and 3D projection with JAX on large scenes, high VRAM is recommended, otherwise if CUDA OOM error, decrease:
 
 ```
 val
@@ -180,7 +181,6 @@ val
 The input data is structured by geographic zone, with RGB images, semantic masks, LiDAR scans, and camera pose files.
 The structure of the GridNet-HD dataset remains the same (see [GridNet-HD dataset](https://huggingface.co/datasets/heig-vd-geo/GridNet-HD) for more information)
 
-
 ---
 
 ## Setup & Installation
@@ -227,28 +227,28 @@ Each mode is selected via the `--mode` argument in `main.py`.
 
 ### Results
 
-The following table summarizes the per-class Intersection over Union (IoU) scores on the validation and test datasets at 3D level. The model was trained using the configuration specified in `config.yaml`.
-
-| Class                      | IoU (Validation set) (%)| IoU (Test set) (%)|
-|---------------------------|------------------|------------|
-| Pylon                     |  86.43            |   85.09     |
-| Conductor cable           |  58.22            |   64.82     |
-| Structural cable          |  48.84            |   45.06     |
-| Insulator                 |  81.16            |   71.07     |
-| High vegetation           |  90.89            |   83.86     |
-| Low vegetation            |  71.89            |   63.43     |
-| Herbaceous vegetation     |  93.87            |   84.45     |
-| Rock, gravel, soil        |  91.40            |   38.62     |
-| Impervious soil (Road)    |  85.84            |   80.69     |
-| Water                     |  44.14            |   74.87     |
-| Building                  |  86.76            |   68.09     |
-| **Mean IoU (mIoU)**       | **76.31**         | **69.10**   |
+The following table summarizes the per-class Intersection over Union (IoU) scores on the test set at 3D level. The model was trained using the configuration specified in `config.yaml`.
+
+| Class                     | IoU (Test set) (%)|
+|---------------------------|------------|
+| Pylon                     |   85.09     |
+| Conductor cable           |   64.82     |
+| Structural cable          |   45.06     |
+| Insulator                 |   71.07     |
+| High vegetation           |   83.86     |
+| Low vegetation            |   63.43     |
+| Herbaceous vegetation     |   84.45     |
+| Rock, gravel, soil        |   38.62     |
+| Impervious soil (Road)    |   80.69     |
+| Water                     |   74.87     |
+| Building                  |   68.09     |
+| **Mean IoU (mIoU)**       | **69.10**   |
 
 ### Pretrained Weights
 
 🔗 **Pretrained weights** for the best performing model are available for download directly in this repo.
 
-> This checkpoint corresponds to the model trained using the configuration in `config.yaml`, achieving a mean IoU of **76.31%** on the validation set and **69.10%** on test test.
+> This checkpoint corresponds to the model trained using the configuration in `config.yaml`, achieving a mean IoU of **69.10%** on test set.
 
 ---
 
@@ -318,14 +318,13 @@ This project is open-sourced under the MIT License.
 
 ## Contact
 
-For questions, issues, or contributions, please open an issue on the repository or contact the project maintainer.
-
+For questions, issues, or contributions, please open an issue on the repository.
 
 ---
 
 ## Citation
 
-If you use this repo in research, please cite: (TODO export arxiv preprint)
+If you use this repo in research, please cite:
 
     GridNet-HD: A High-Resolution Multi-Modal Dataset for LiDAR-Image Fusion on Power Line Infrastructure
     Masked Authors

config.yaml

@@ -2,7 +2,7 @@ data:
   # Root folder containing your sub-folders (t1z4, t2z5, etc.)
   root_dir: "/path/to/GridNet-HD"
   # JSON split file listing train/val/test folders
-  split_file: "/path/to/split.json"
+  split_file: "/path/to/GridNet-HD/split.json"
   # First resize each image+mask
   resize_size: [1760, 1318] # PIL style (width, height)
   # Then random-crop (train) or sliding-window (val/test) to this size (H, W)
@@ -54,7 +54,6 @@ model:
 training:
   # Where to save checkpoints & logs
   output_dir: "./outputs/run"
-  # Random seed for reproducibility
   seed: 42
   # Batch size for training
   batch_size: 32
@@ -73,7 +72,7 @@ training:
 val:
  batch_size: 8 # number of images per batch
  num_workers: 8
- batch_size_proj: 20000000 # number of points per batch to project on images
+ batch_size_proj: 5000000 # number of points per batch to project on images
  
 
 # =============================================================================

main.py

@@ -153,7 +153,7 @@ def main():
         model.load_state_dict(torch.load(args.weights_path))
         inference(model, val_loader, device, ds_args["crop_size"], ds_args["crop_size"],
                   out_dir / "predictions")
-
+        logging.info(f"Inference Image complete. Predictions saved to {out_dir/'predictions'}")
         conf_mat = np.zeros((cfg["model"]["num_classes"], cfg["model"]["num_classes"]), dtype=int)
         for zone in sorted(os.listdir(out_dir / "predictions")):
             output_las_path = out_dir / "predictions" / zone / f"{zone}_with_classif.las"

requirements.txt

@@ -10,6 +10,8 @@ GitPython==3.1.44
 huggingface-hub==0.30.2
 idna==3.10
 jax==0.6.0
+jax-cuda12-pjrt==0.6.0
+jax-cuda12-plugin==0.6.0
 jaxlib==0.6.0
 Jinja2==3.1.6
 joblib==1.5.0
@@ -21,9 +23,10 @@ networkx==3.4.2
 numpy==2.2.5
 nvidia-cublas-cu12==12.6.4.1
 nvidia-cuda-cupti-cu12==12.6.80
+nvidia-cuda-nvcc-cu12==12.9.41
 nvidia-cuda-nvrtc-cu12==12.6.77
 nvidia-cuda-runtime-cu12==12.6.77
-nvidia-cudnn-cu12==9.5.1.17
+nvidia-cudnn-cu12==9.10.0.56
 nvidia-cufft-cu12==11.3.0.4
 nvidia-cufile-cu12==1.11.1.6
 nvidia-curand-cu12==10.3.7.77