Adding example code to run on Modal.

modal/README.md

@@ -0,0 +1,29 @@
+# OpenPCDet on Modal
+
+This is a guide to running PCDet on [Modal](https://modal.com/), a platform for running machine learning workflows on the cloud.
+
+In this example we will train a pointpillar model on the nuScenes mini dataset. Credit to [ies0411](https://github.com/ies0411) who contributed the dockerfile in https://github.com/open-mmlab/OpenPCDet/pull/1513.
+
+## Install modal
+
+* `pip install modal`
+* `modal setup`
+
+See https://modal.com/docs/guide for more information.
+
+## Prepare data
+
+* Download the nuScenes mini dataset
+* Create shared Modal volume: `modal volume create nuscenes`.
+* Optionally: delete all image and radar folders to speed up the next step.
+* Upload data to the volume: `cd path-to-downloaded-data`, `modal volume put nuscenes v1.0-mini`.
+
+## Format the data
+
+Build the nuScenes `infos` used by OpenPCDet. This is a one-time operation.
+
+* `modal run setup_nuscenes.py`
+
+## Train the model
+
+* `modal run train_pp.py`

modal/current.Dockerfile

@@ -0,0 +1,23 @@
+FROM nvidia/cuda:11.6.2-devel-ubuntu20.04
+ENV DEBIAN_FRONTEND=noninteractive
+
+# Get all dependencies
+RUN apt-get update && apt-get install -y \
+    git zip build-essential cmake libssl-dev python3-dev  python3-pip python3-pip cmake ninja-build git wget ca-certificates ffmpeg libsm6 libxext6 &&\
+    rm -rf /var/lib/apt/lists/*
+
+RUN ln -sv /usr/bin/python3 /usr/bin/python
+ENV PATH="/root/.local/bin:${PATH}"
+RUN wget -O /root/get-pip.py https://bootstrap.pypa.io/get-pip.py && python3 /root/get-pip.py --user
+
+# PyTorch for CUDA 11.6
+RUN pip install torch==1.13.1+cu116 torchvision==0.14.1+cu116 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu116
+ENV TORCH_CUDA_ARCH_LIST="3.5;5.0;6.0;6.1;7.0;7.5;8.0;8.6+PTX;Kepler;Kepler+Tesla;Maxwell;Maxwell+Tegra;Pascal;Volta;Turing;8.6"
+
+RUN pip install numpy==1.23.0 llvmlite numba opencv-python tensorboardX easydict pyyaml scikit-image tqdm SharedArray open3d mayavi av2 pyquaternion kornia==0.6.8 nuscenes-devkit==1.0.5 spconv-cu116
+RUN python -m pip install --user jupyter
+RUN git clone https://github.com/open-mmlab/OpenPCDet.git
+WORKDIR OpenPCDet
+RUN python setup.py develop
+RUN pip install torch-scatter==2.0.9 -f https://data.pyg.org/whl/torch-1.13.1+cu116.html
+

modal/setup_nuscenes.py

@@ -0,0 +1,18 @@
+from shared import pcdet_cuda_image, volumes
+import modal
+
+app = modal.App("pcdet")
+
+@app.function(image=pcdet_cuda_image, gpu="T4", timeout=3600*2, volumes=volumes)
+def build_nuscenes():
+    print("-> Building nuScenes")
+
+    import subprocess
+    subprocess.run(["python", "-m", "pcdet.datasets.nuscenes.nuscenes_dataset", "--func", "create_nuscenes_infos", 
+    "--cfg_file", "tools/cfgs/dataset_configs/nuscenes_dataset.yaml", "--version", "v1.0-mini"])
+
+    print("-> Done building nuScenes")
+
+@app.local_entrypoint()
+def main():
+    build_nuscenes.remote()

modal/shared.py

@@ -0,0 +1,11 @@
+import modal
+
+pcdet_cuda_image = (
+    modal.Image.from_dockerfile("current.dockerfile")
+    .copy_local_dir("../tools", "/OpenPCDet/tools")
+)
+
+volume = modal.Volume.from_name("nuscenes")
+VOL_MOUNT_PATH = "/OpenPCDet/data/nuscenes"
+
+volumes={VOL_MOUNT_PATH: volume}

modal/train_pp.py

@@ -0,0 +1,24 @@
+import os
+from shared import pcdet_cuda_image, volumes
+import modal
+
+app = modal.App("pcdet")
+
+@app.function(image=pcdet_cuda_image, 
+              volumes=volumes,
+              cpu=8.0, 
+              memory=32768, 
+              gpu="T4", 
+              timeout=3600*2)
+def train_pointpillars():
+    print("-> Training PP")
+
+    import subprocess
+    os.chdir("/OpenPCDet/tools")
+    subprocess.run(["python", "train.py", "--cfg_file", "cfgs/nuscenes_models/cbgs_pp_multihead.yaml"])
+
+    print("-> Done training PP")
+
+@app.local_entrypoint()
+def main():
+    train_pointpillars.remote()

pcdet/datasets/__init__.py

@@ -12,7 +12,6 @@
 from .pandaset.pandaset_dataset import PandasetDataset
 from .lyft.lyft_dataset import LyftDataset
 from .once.once_dataset import ONCEDataset
-from .argo2.argo2_dataset import Argo2Dataset
 from .custom.custom_dataset import CustomDataset
 
 __all__ = {
@@ -24,9 +23,20 @@
     'LyftDataset': LyftDataset,
     'ONCEDataset': ONCEDataset,
     'CustomDataset': CustomDataset,
-    'Argo2Dataset': Argo2Dataset
 }
 
+def DataSetClassFactory(name):
+    if name in __all__:
+        return __all__[name]
+    else:
+        # argo2_dataset.py does a global import of av2, 
+        # which is not part of the standard install.
+        # This is a workaround to avoid requiring av2 to be installed 
+        # unless the user is using the Argo2Dataset.
+        from .argo2.argo2_dataset import Argo2Dataset
+        return Argo2Dataset
+
+
 
 class DistributedSampler(_DistributedSampler):
 
@@ -54,7 +64,7 @@ def __iter__(self):
 def build_dataloader(dataset_cfg, class_names, batch_size, dist, root_path=None, workers=4, seed=None,
                      logger=None, training=True, merge_all_iters_to_one_epoch=False, total_epochs=0):
 
-    dataset = __all__[dataset_cfg.DATASET](
+    dataset = DataSetClassFactory(dataset_cfg.DATASET)(
         dataset_cfg=dataset_cfg,
         class_names=class_names,
         root_path=root_path,

tools/cfgs/nuscenes_models/cbgs_pp_multihead.yaml

@@ -5,6 +5,7 @@ DATA_CONFIG:
     _BASE_CONFIG_: cfgs/dataset_configs/nuscenes_dataset.yaml
 
     POINT_CLOUD_RANGE: [-51.2, -51.2, -5.0, 51.2, 51.2, 3.0]
+    VERSION: 'v1.0-mini'
     DATA_PROCESSOR:
         -   NAME: mask_points_and_boxes_outside_range
             REMOVE_OUTSIDE_BOXES: True