This repository is forked from the official implementation SMOKE: Single-Stage Monocular 3D Object Detection via Keypoint Estimation with some updates. Private function only.

Resource

The pretrained weights can be downloaded here.

Requirements

All codes are tested under the following environment:

• Ubuntu 16.04
• Python 3.8
• Pytorch 1.8.1
• CUDA 11.1

Dataset

We train and test our model on official KITTI 3D Object Dataset. Please first download the dataset and organize it as following structure:

kitti
│──ImageSets
│──training
│    ├──calib
│    ├──label_2 
│    └──image_2
└──testing
     ├──calib
     └──image_2

Install

git clone [email protected]:excitohe/smoke.git

python setup.py develop

mkdir datasets

ln -s /path_to_kitti_dataset datasets/kitti

Getting started

First check the config file under configs/.

We train the model on 4 GPUs with 32 batch size:

python tools/plain_train_net.py --num-gpus 4 --config-file "configs/smoke_dla34_gn.yaml"

For single GPU training, simply run:

python tools/plain_train_net.py --config-file "configs/smoke_dla34_gn.yaml"

We currently only support single GPU testing:

python tools/plain_train_net.py --eval-only --config-file "configs/smoke_dla34_gn.yaml"