zhongxiayan/squeezedet

A tensorflow implementation for SqueezeDet, a convolutional neural network for object detection.

Ternarizing SqueezeDet

We attempt ternarization on layers of SqueezeDet. See below for the paper on SqueezeDet. We cloned the repo from https://github.com/BichenWuUCB/squeezeDet and made modifications.

SqueezeDet: Unified, Small, Low Power Fully Convolutional Neural Networks for Real-Time Object Detection for Autonomous Driving

By Bichen Wu, Forrest Iandola, Peter H. Jin, Kurt Keutzer (UC Berkeley & DeepScale)

This repository contains a tensorflow implementation of SqueezeDet, a convolutional neural network based object detector described in our paper: https://arxiv.org/abs/1612.01051. If you find this work useful for your research, please consider citing:

@inproceedings{squeezedet,
    Author = {Bichen Wu and Forrest Iandola and Peter H. Jin and Kurt Keutzer},
    Title = {SqueezeDet: Unified, Small, Low Power Fully Convolutional Neural Networks for Real-Time Object Detection for Autonomous Driving},
    Journal = {arXiv:1612.01051},
    Year = {2016}
}

Installation

The following instructions are written for Linux-based distros.

• Clone the SqueezeDet repository:

  git clone https://github.com/ZhongxiaYan/squeezeDet.git

  Let's call the top level directory of SqueezeDet $SQDT_ROOT.

Virtual Environment and Demo

Optional, refer to https://github.com/BichenWuUCB/squeezeDet for details.

Download Data and Pretrained Model

• Download KITTI object detection dataset: images and labels. Put them under $SQDT_ROOT/data/KITTI/. Unzip them, then you will get two directories: $SQDT_ROOT/data/KITTI/training/ and $SQDT_ROOT/data/KITTI/testing/.

• Now we need to split the training data into a training set and a vlidation set.

  cd $SQDT_ROOT/data/KITTI/
  mkdir ImageSets
  cd ./ImageSets
  ls ../training/image_2/ | grep ".png" | sed s/.png// > trainval.txt

  trainval.txt contains indices to all the images in the training data. In our experiments, we randomly split half of indices in trainval.txt into train.txt to form a training set and rest of them into val.txt to form a validation set. For your convenience, we provide a script to split the train-val set automatically. Simply run

  cd $SQDT_ROOT/data/
  python random_split_train_val.py

  then you should get the train.txt and val.txt under $SQDT_ROOT/data/KITTI/ImageSets.

  When above two steps are finished, the structure of $SQDT_ROOT/data/KITTI/ should at least contain:

  $SQDT_ROOT/data/KITTI/
                    |->training/
                    |     |-> image_2/00****.png
                    |     L-> label_2/00****.txt
                    |->testing/
                    |     L-> image_2/00****.png
                    L->ImageSets/
                          |-> trainval.txt
                          |-> train.txt
                          L-> val.txt

• Next, download the SqueezeNet CNN model pretrained for ImageNet classification:

  cd $SQDT_ROOT/data/
  wget https://www.dropbox.com/s/fzvtkc42hu3xw47/SqueezeNet.tgz
  tar -xzvf SqueezeNet.tgz

• Compile the official evaluation script of KITTI dataset

  cd $SQDT_ROOT/src/dataset/kitti-eval
  make

Defining a Model

Let's call our model $model. We define the model architecture in $SQDT_ROOT/models/$model/load_model.py. The model is a class that is returned by the function load_model, and the class must contain member variables used in $SQDT_ROOT/src/run.py. Refer to $SQDT_ROOT/models/squeezedet_plus/load_model.py as an example.

Defining a Config

We can define a specific set of hyperparameters (name this set $config) for $model in $SQDT_ROOT/models/$model/$config/config.json. The model may load in and use these defined configs. Refer to the default config for squeezedet_plus model as an example.

Training

We train a model with a specific set of configurations at a time.

cd $SQDT_ROOT/
python src/run.py models/$model/$config/

Training logs and checkpoints are saved to the directory python src/run.py models/$model/$config/train .

Evaluation

You can evaluate a model with a set of configs by running

cd $SQDT_ROOT/
python src/run.py models/$model/$config/ --train False

Evaluation logs and checkpoints are saved to the directory python src/run.py models/$model/$config/val .

You can run evaluation simultaneously with training for a GPU with 8G+ memory.

Finally, to monitor training and evaluation process, you can use tensorboard by

tensorboard --logdir=models/$model/$config/