Skip to content

Latest commit

 

History

History
117 lines (87 loc) · 5.91 KB

ONLINE_RPN_EXP.md

File metadata and controls

117 lines (87 loc) · 5.91 KB
Raw

You can run the main experiments reported in the work Fast Region Proposal Learning for Object Detection for Robotics by following the reported instructions.

Code installation

Please follow the instructions for installations reported at this link.

Datasets Download

Depending on the experiment that you want to run, you need to download the required dataset. Namely, if you want to run the experiment on the 30 objects identification task from the iCubWorld Transformations dataset (iCWT), you need to download such dataset, following the subsequent instructions.

cd $HOME_DIR/Data/datasets/iCWT/iCubWorld-Transformations
mkdir Images
cd Images
wget https://zenodo.org/record/835510/files/part1.tar.gz
tar -xvf part1.tar.gz --strip 1
wget https://zenodo.org/record/835510/files/part2.tar.gz
tar -xvf part2.tar.gz --strip 1
wget https://zenodo.org/record/835510/files/part3.tar.gz
tar -xvf part3.tar.gz --strip 1
wget https://zenodo.org/record/835510/files/part4.tar.gz
tar -xvf part4.tar.gz --strip 1

rm part1.tar.gz
rm part2.tar.gz
rm part3.tar.gz
rm part4.tar.gz

cd $HOME_DIR/Data/datasets/iCWT/iCubWorld-Transformations
mkdir Annotations
cd Annotations
wget https://zenodo.org/record/1227305/files/Annotations_refined.tar.gz
tar -xvf Annotations_refined.tar.gz --strip 1
rm Annotations_refined.tar.gz

cd $HOME_DIR/Data/datasets/iCWT/iCubWorld-Transformations_manual
ln -s $HOME_DIR/Data/datasets/iCWT/iCubWorld-Transformations/Images .
mkdir Annotations
cd Annotations
wget https://zenodo.org/record/2563223/files/Annotations_manual.tar.gz
tar -xvf Annotations_manual.tar.gz --strip 1
rm Annotations_manual.tar.gz

Similarly, if you want to run the experiment on the 21 objects identification task from the TABLE-TOP dataset, you need to download such dataset, following the subsequent instructions.

cd $HOME_DIR/Data/datasets/iCWT/TABLE-TOP
mkdir Annotations
cd Annotations
wget https://zenodo.org/record/3970624/files/table_top_annotations.tar.xz
tar -xvf table_top_annotations.tar.xz --strip 1
rm table_top_annotations.tar.xz

cd $HOME_DIR/Data/datasets/iCWT/TABLE-TOP
mkdir Images
cd Images
wget https://zenodo.org/record/3970624/files/table_top_images.tar.xz
tar -xvf table_top_images.tar.xz --strip 1
rm table_top_images.tar.xz

Pretrained Feature Extractors Download

Create the pretrained_feature_extractors directory, where you will download the pretrained feature extractors that you need to run the experiments.

cd $HOME_DIR/Data
mkdir pretrained_feature_extractors
cd pretrained_feature_extractors

If you want to run the experiment on the 30 objects identification task from the iCubWorld Transformations dataset (iCWT), you need to download the feature extractor with the following command.

 wget --content-disposition https://istitutoitalianotecnologia-my.sharepoint.com/:u:/g/personal/federico_ceola_iit_it/EW4b1wC905JDgKkrPG6nmmABAlDNqtbiKdnQN7QN0NoO6A?download=1

Similarly, if you want to run the experiment on the 21 objects identification task from the TABLE-TOP dataset, you need to download the feature extractor with the following command.

wget --content-disposition https://istitutoitalianotecnologia-my.sharepoint.com/:u:/g/personal/federico_ceola_iit_it/EY1Bu31xC_xCviY0qxdPwKIBBRlQfjSJ_MX1xXs97AVH1A?download=1

Note: at the end of the installation, remember to unset HOME_DIR

Usage

By modifying the configuration files in the experiments/configs folder and by substituting the files with the proper ones for your data, this code allows you to run customized experiments. However, in this repository we provide you with the configuration files and scripts that are required to reproduce the main experiments in the presented paper. For all the experiments that you want to reproduce, you have to run the script experiments/run_experiment.py and to properly set command line arguments. Some examples will be provided in the two following subsections. For additional information please refer to the helper of the script, running the command python run_experiment.py -h in the experiments directory.

In the experiments/configs you can find four categories of configuration files:

  • config_rpn_experiment_name.yaml: sets parameters for RPN's feature extraction.
  • config_detector_experiment_name.yaml: sets parameters for detector's feature extraction.
  • config_online_detection_experiment_name.yaml: sets parameters for the O-OD experiment (3rd row of Table 1 and Table 2 in the paper).
  • config_online_online_rpn_detection_experiment_name.yaml: sets parameters for the Ours experiment (5th row of Table 1 and Table 2 in the paper).

Important: if you have more than one GPU available, before running an experiment, you have to set the number of the GPU that you want to use (only one) with the command export CUDA_VISIBLE_DEVICES=number_of_the_gpu

iCWT: different objects, same setting.

To reproduce the results of the O-OD experiment on the iCWT dataset reported in the Table 1 of the paper, in the experiments folder you have to run the command

python run_experiment.py --icwt30 --only_ood.

To reproduce Ours experiment you have to run the command

python run_experiment.py --icwt30.

If you do not set from command line an output directory, experiment's results will be saved in the experiments/icwt30_experiment/result.txt file.

TABLE-TOP: different objects and setting.

Similarly to the iCWT experiment, to reproduce the results of the O-OD experiment on the TABLE-TOP dataset reported in the Table 2 of the paper, in the experiments folder you have to run the command

python run_experiment.py --only_ood.

To reproduce Ours experiment you have to run the command

python run_experiment.py.

If you do not set from command line an output directory, experiment's results will be saved in the experiments/tabletop_experiment/result.txt file.