Skip to content

Latest commit

 

History

History

semantic

README.md

3D-MiniNet: Pytorch Implementation

This code allows to reproduce the experiments from the 3D-MiniNet paper.

Dependencies

First you need to install the nvidia driver and CUDA, so have fun!

  • CUDA Installation guide: link

  • System dependencies:

    $ sudo apt-get update 
$ sudo apt-get install -yqq  build-essential ninja-build \
  python3-dev python3-pip apt-utils curl git cmake unzip autoconf autogen \
  libtool mlocate zlib1g-dev python3-numpy python3-wheel wget \
  software-properties-common openjdk-8-jdk libpng-dev  \
  libxft-dev ffmpeg python3-pyqt5.qtopengl
$ sudo updatedb

  • Python dependencies

    $ sudo pip3 install -r requirements.txt

Dataset

Download the SemanticKitti dataset here

Configuration files

Architecture configuration files are located at config/arch Dataset configuration files are located at config/labels

Apps

ALL SCRIPTS CAN BE INVOKED WITH -h TO GET EXTRA HELP ON HOW TO RUN THEM

Visualization

To visualize the data (in this example sequence 00):

$ ./visualize.py -d /path/to/dataset/ -s 00

To visualize the predictions (in this example sequence 00):

$ ./visualize.py -d /path/to/dataset/ -p /path/to/predictions/ -s 00

Training

To train a network (from scratch):

$ ./train.py -d /path/to/dataset -ac config/arch/CHOICE.yaml -l /path/to/log

To train a network (from pretrained model):

$ ./train.py -d /path/to/dataset -ac config/arch/CHOICE.yaml -dc config/labels/CHOICE.yaml -l /path/to/log -p /path/to/pretrained

Change any training parameters like the number of training epochs in the config/labels/CHOICE.yaml configuration file.

Besides, for a minor boost of performance you can:

  • Play with the hyperparameters
  • Train more epochs or doing several trainings lowering the learning rate
  • Change the data augmentation
  • Add the validation set to the training: line 56 of the (modules/trainer.py): train_sequences=self.DATA["split"]["train"] + self.DATA["split"]["valid"],

Inference

To infer the predictions for the entire dataset:

$ ./infer.py -d /path/to/dataset/ -l /path/for/predictions -m /path/to/model

Evaluation

Upload test inferences on the evaluation platform

For validating against the validation set:

$ ./evaluate_iou.py -d /path/to/dataset/ -p /path/for/predictions --split valid

Pre-trained Models

These models have been trained both on train and validation data.

To enable kNN post-processing, just change the boolean value to True in the arch_cfg.yaml file parameter, inside the model directory.

Troubleshooting

  • Please, **if you have a CUDA out memory error lower the batch size, although performance can be compromised.

  • If you have some problems with the GPU (like you have only 1 gpu and it is not founding it correctly) you may add at the beginning of the training file:

n_gpu = 0
os.environ["CUDA_VISIBLE_DEVICES"] = str(n_gpu)