'add downloading of scannet'

.devcontainer/Dockerfile

@@ -35,5 +35,11 @@ RUN pip3 install setuptools poetry
 COPY pyproject.toml .
 COPY poetry.lock .
 
-RUN POETRY_VIRTUALENVS_CREATE=false poetry install
-RUN pre-commit install
+RUN poetry install --no-root
+RUN poetry env info --path > /python_path
+RUN cat /python_path
+
+RUN git clone https://github.com/StanfordVL/MinkowskiEngine.git /tmp/ME \
+    && cd /tmp/ME \
+    && export VENV=`cat /python_path` \
+    && $VENV/bin/python setup.py install --cpu_only

.gitignore

@@ -19,3 +19,4 @@ measurements/*.pickle
 _build
 docs_old
 /test/kernels/dispositions
+*.egg-info*

README.md

@@ -8,6 +8,69 @@ This is a framework for running common deep learning models for point cloud anal
 
 The framework allows lean and yet complex model to be built with minimum effort and great reproducibility.
 
+## Project structure
+
+```bash
+├─ benchmark               # Output from various benchmark runs
+├─ conf                    # All configurations for training nad evaluation leave there
+├─ dashboard               # A collection of notebooks that allow result exploration and network debugging
+├─ docker                  # Docker image that can be used for inference or training
+├─ docs                    # All the doc
+├─ eval.py                 # Eval script
+├─ find_neighbour_dist.py  # Script to find optimal #neighbours within neighbour search operations
+├─ forward_scripts         # Script that runs a forward pass on possibly non annotated data
+├─ outputs                 # All outputs from your runs sorted by date
+├─ scripts                 # Some scripts to help manage the project
+├─ src  
+    ├─ core                # Core components
+    ├─ datasets            # All code related to datasets
+    ├─ metrics             # All metrics and trackers
+    ├─ models              # All models
+    ├─ modules             # Basic modules that can be used in a modular way
+    ├─ utils               # Various utils
+    └─ visualization       # Visualization
+├─ test
+└─ train.py                # Main script to launch a training
+```
+
+As a general philosophy we have split datasets and models by task. For example, datasets has three subfolders:
+
+- segmentation
+- classification
+- registration
+
+where each folder contains the dataset related to each task.
+
+## Methods currently implemented
+
+* **[PointNet](https://github.com/nicolas-chaulet/deeppointcloud-benchmarks/blob/master/src/modules/PointNet/modules.py#L54)** from Charles R. Qi *et al.*: [PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation](https://arxiv.org/abs/1612.00593) (CVPR 2017)
+* **[PointNet++](https://github.com/nicolas-chaulet/deeppointcloud-benchmarks/tree/master/src/modules/pointnet2)** from Charles from Charles R. Qi *et al.*: [PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space](https://arxiv.org/abs/1706.02413)
+* **[RSConv](https://github.com/nicolas-chaulet/deeppointcloud-benchmarks/tree/master/src/modules/RSConv)** from Yongcheng Liu *et al.*: [Relation-Shape Convolutional Neural Network for Point Cloud Analysis](https://arxiv.org/abs/1904.07601) (CVPR 2019)
+* **[RandLA-Net](https://github.com/nicolas-chaulet/deeppointcloud-benchmarks/tree/master/src/modules/RandLANet)** from Qingyong Hu *et al.*: [RandLA-Net: Efficient Semantic Segmentation of Large-Scale Point Clouds](https://arxiv.org/abs/1911.11236)
+* **[PointCNN](https://github.com/nicolas-chaulet/deeppointcloud-benchmarks/tree/master/src/modules/PointCNN)** from Yangyan Li *et al.*: [PointCNN: Convolution On X-Transformed Points](https://arxiv.org/abs/1801.07791) (NIPS 2018)
+* **[KPConv](https://github.com/nicolas-chaulet/deeppointcloud-benchmarks/tree/master/src/modules/KPConv)** from Hugues Thomas *et al.*: [KPConv: Flexible and Deformable Convolution for Point Clouds](https://arxiv.org/abs/1801.07791) (ICCV 2019)
+* **[MinkowskiEngine](https://github.com/nicolas-chaulet/deeppointcloud-benchmarks/tree/master/src/modules/MinkowskiEngine)** from Christopher Choy *et al.*: [4D Spatio-Temporal ConvNets: Minkowski Convolutional Neural Networks](https://arxiv.org/abs/1904.08755) (CVPR'19)
+
+
+## Available datasets
+### Segmentation
+* **[Scannet](https://github.com/ScanNet/ScanNet)** from Angela Dai *et al.*: [ScanNet: Richly-annotated 3D Reconstructions of Indoor Scenes](https://arxiv.org/abs/1702.04405)
+
+* **[S3DIS](http://buildingparser.stanford.edu/dataset.html)** from Iro Armeni *et al.*: [Joint 2D-3D-Semantic Data for Indoor Scene Understanding](https://arxiv.org/abs/1702.01105)
+```
+* S3DIS 1x1
+* S3DIS Room
+* S3DIS Fused
+```
+
+* **[Shapenet](https://www.shapenet.org/)** from Iro Armeni *et al.*: [ShapeNet: An Information-Rich 3D Model Repository](https://arxiv.org/abs/1512.03012)
+
+### Registration
+* **[3DMatch](http://3dmatch.cs.princeton.edu)** from Andy Zeng *et al.*: [3DMatch: Learning Local Geometric Descriptors from RGB-D Reconstructions](https://arxiv.org/abs/1603.08182)
+
+### Classification
+* **[ModelNet](https://modelnet.cs.princeton.edu)** from Zhirong Wu *et al.*: [3D ShapeNets: A Deep Representation for Volumetric Shapes](https://people.csail.mit.edu/khosla/papers/cvpr2015_wu.pdf)
+
 ## Getting started
 ### Requirements:
 * CUDA > 10

conf/data/segmentation/scannet.yaml

@@ -0,0 +1,39 @@
+data:
+    class: scannet.ScannetDataset
+    task: segmentation
+    dataroot: data
+    grid_size: 0.05
+    version: 'v2'
+    use_instance_labels: False
+    use_instance_bboxes: False
+    donotcare_class_ids: []
+    max_num_point: None
+    process_workers: 1
+    train_transform:
+      - transform: ToSparseInput
+        params:
+          grid_size: ${data.grid_size}
+          mode: "mean"
+          quantizing_func: "round"
+      - transform: AddOnes
+      - transform: AddFeatsByKeys
+        params:
+          list_add_to_x: [True,True]
+          feat_names: ["ones","rgb"]
+          input_nc_feats: [1,3]
+          stricts: [True, True]
+          delete_feats: [True, True]
+    val_transform:
+      - transform: ToSparseInput
+        params:
+          grid_size: ${data.grid_size}
+          mode: "mean"
+          quantizing_func: "round"
+      - transform: AddOnes
+      - transform: AddFeatsByKeys
+        params:
+          list_add_to_x: [True,True]
+          feat_names: ["ones","rgb"]
+          input_nc_feats: [1,3]
+          stricts: [True, True]
+          delete_feats: [True, True]

conf/models/segmentation/minkowski_baseline.yaml

@@ -58,4 +58,100 @@ models:
         class: minkowski.Minkowski_Baseline_Model
         conv_type: "SPARSE"
         model_name: "MinkUNet34C"
-        D: 3
+        D: 3
+
+    Res16UNet14:
+        class: minkowski.Minkowski_Baseline_Model
+        conv_type: "SPARSE"
+        model_name: "Res16UNet14"
+        D: 3
+
+    Res16UNet18:
+        class: minkowski.Minkowski_Baseline_Model
+        conv_type: "SPARSE"
+        model_name: "Res16UNet18"
+        D: 3
+
+    Res16UNet34:
+        class: minkowski.Minkowski_Baseline_Model
+        conv_type: "SPARSE"
+        model_name: "Res16UNet34"
+        D: 3
+
+    Res16UNet14A:
+        class: minkowski.Minkowski_Baseline_Model
+        conv_type: "SPARSE"
+        model_name: "Res16UNet14A"
+        D: 3  
+
+    Res16UNet14A2:
+        class: minkowski.Minkowski_Baseline_Model
+        conv_type: "SPARSE"
+        model_name: "Res16UNet14A2"
+        D: 3   
+
+    Res16UNet14B:
+        class: minkowski.Minkowski_Baseline_Model
+        conv_type: "SPARSE"
+        model_name: "Res16UNet14B"
+        D: 3  
+
+    Res16UNet14B2:
+        class: minkowski.Minkowski_Baseline_Model
+        conv_type: "SPARSE"
+        model_name: "Res16UNet14B2"
+        D: 3 
+
+    Res16UNet14B3:
+        class: minkowski.Minkowski_Baseline_Model
+        conv_type: "SPARSE"
+        model_name: "Res16UNet14B3"
+        D: 3 
+
+    Res16UNet14C:
+        class: minkowski.Minkowski_Baseline_Model
+        conv_type: "SPARSE"
+        model_name: "Res16UNet14C"
+        D: 3  
+
+    Res16UNet14D:
+        class: minkowski.Minkowski_Baseline_Model
+        conv_type: "SPARSE"
+        model_name: "Res16UNet14D"
+        D: 3  
+
+    Res16UNet18A:
+        class: minkowski.Minkowski_Baseline_Model
+        conv_type: "SPARSE"
+        model_name: "Res16UNet18A"
+        D: 3  
+
+    Res16UNet18B:
+        class: minkowski.Minkowski_Baseline_Model
+        conv_type: "SPARSE"
+        model_name: "Res16UNet18B"
+        D: 3  
+
+    Res16UNet18D:
+        class: minkowski.Minkowski_Baseline_Model
+        conv_type: "SPARSE"
+        model_name: "Res16UNet18D"
+        D: 3  
+
+    Res16UNet34A:
+        class: minkowski.Minkowski_Baseline_Model
+        conv_type: "SPARSE"
+        model_name: "Res16UNet34A"
+        D: 3  
+
+    Res16UNet34B:
+        class: minkowski.Minkowski_Baseline_Model
+        conv_type: "SPARSE"
+        model_name: "Res16UNet34B"
+        D: 3   
+
+    Res16UNet34C:
+        class: minkowski.Minkowski_Baseline_Model
+        conv_type: "SPARSE"
+        model_name: "Res16UNet34C"
+        D: 3 

conf/training/minkowski_scannet.yaml

@@ -0,0 +1,44 @@
+# Ref: https://github.com/chrischoy/SpatioTemporalSegmentation/blob/master/config.py
+training:
+    epochs: 100
+    num_workers: 4
+    batch_size: 16
+    shuffle: True
+    cuda: 1
+    precompute_multi_scale: False # Compute multiscate features on cpu for faster training / inference
+    optim:
+        base_lr: 0.01
+        # accumulated_gradient: -1 # Accumulate gradient accumulated_gradient * batch_size
+        grad_clip: -1
+        optimizer:
+            class: SGD
+            params:
+                lr: ${training.optim.base_lr} # The path is cut from training
+                momentum: 0.9
+                dampening: 0.1
+                weight_decay: 1e-4
+        lr_scheduler: ${lr_scheduler}
+        bn_scheduler:
+            bn_policy: "step_decay"
+            params:
+                bn_momentum: 0.02
+                bn_decay: 1
+                decay_step : 10
+                bn_clip : 1e-2
+    weight_name: "latest" # Used during resume, select with model to load from [miou, macc, acc..., latest]
+    enable_cudnn: True
+    checkpoint_dir: ""
+
+# Those arguments within experiment defines which model, dataset and task to be created for benchmarking
+# parameters for Weights and Biases
+wandb:
+    entity: ""
+    project: scannet
+    log: False
+    notes:
+    name:
+    public: True # It will be display the model within wandb log, else not.
+
+    # parameters for TensorBoard Visualization
+tensorboard:
+    log: True

docker/Dockerfile

@@ -18,7 +18,11 @@ RUN pip3 install setuptools poetry
 COPY pyproject.toml .
 COPY poetry.lock .
 
-RUN POETRY_VIRTUALENVS_CREATE=false poetry install -E MinkowskiEngine --no-dev --no-root && rm -rf /root/.cache
+RUN POETRY_VIRTUALENVS_CREATE=false poetry install --no-dev --no-root && rm -rf /root/.cache
+
+RUN git clone https://github.com/StanfordVL/MinkowskiEngine.git /tmp/ME \
+    && cd /tmp/ME \
+    && python3 setup.py install --cpu_only
 
 ARG MODEL=""
 ENV WORKDIR=/dpb

pyproject.toml

@@ -3,9 +3,6 @@ name = "deeppointcloud_benchmark"
 version = "0.1.0"
 description = ""
 authors = ["Nicolas <nicolas.chaulet@gmail.com>"]
-packages = [
-    { include = "src" }
-]
 
 [tool.poetry.dependencies]
 python = "^3.6"
@@ -40,6 +37,7 @@ param = "^1.9.3"
 codecov = "^2.0.16"
 sphinx = "^2.4.4"
 sphinx-autobuild = "^0.7.1"
+gpustat = "^0.6.0"
 
 [tool.poetry.extras]
 MinkowskiEngine = ["MinkowskiEngine"]