Epic_Kitchens_Object_Detector_Detectron
WIP for Detectron2 and Python3. For the previous working version using python2 and detectron (of the object feature extractor only), refer to the python2 branch.
Detect Objects in the Epic Kitchens dataset using Faster-RCNN as the backbone and the Detectron library.
Once the updates are up, the python2 version will be deprecated.
Implement batch and multi-gpu processing in inference script. --Done
Create pascal-voc evaluator for Epic Kitchens -- In Progress
Implement tester/inference script with detectron2 and python3. --Done
Provide conda environment file or requirements file to setup python environment. -- Done
Modify docker scripts to accomodate updated versions of python and detectron (Currently not working)
Create the feature extractor module with detectron2
Clone the repo and set it up in your local drive.
git clone https://github.com/tridivb/Epic_Kitchens_Object_Detector_Detectron.git
Nvidia Graphics Drivers compatible with Cuda 10
Python >= 3.6
Pytorch >= 1.3
Detectron2
The object frames of the Epic Kitchens Dataset should be downloaded and the folder hierarchy should be in the following way, whis is also the default hierarchy of the dataset:
|---<path to epic kitchens> | |---annotations | | |---EPIC_test_s1_object_video_list.csv | | |---EPIC_test_s2_object_video_list.csv | | |---EPIC_train_object_labels.csv | | |---. | | |---. | |---EPIC_KITCHENS_2018 | | |---object_detection_images | | | |---rgb | | | | |---train | | | | | |---P01 | | | | | | |---P01_01 | | | | | | | |---frame0000000001.jpg | | | | | | | |---. | | | | | | | |---. | | | | | | |---. | | | | | | |---. | | | | |---test | | | | | |---P02 | | | | | | |---P02_12 | | | | | | | |---frame0000000001.jpg | | | | | | | |---. | | | | | | | |---. | | | | | | |---. | | | | | | |---.
Navigate to the Epic_Kitchens_Object_Detector_Detectron directory
cd <path to repo>/Epic_Kitchens_Object_Detector_Detectron
Setup the virtual python environment in your preferred way using virtualenv or conda. Then install the required packages using the requirements.txt file.
pip install -r requirements.txt
Install Detectron2 by following the instructions provided in INSTALL.md
Please note, the docker configuration is not yet ready, so it will not work for now.
Set the paths and parameters in the config/faster_rcnn_R_101_FPN_3x.yaml file.
Alternatively, if you want git to ignore the modified config file, copy it and rename the config file as <new_name>_local.yaml.
cp config/faster_rcnn_R_101_FPN_3x.yaml config/faster_rcnn_R_101_FPN_3x_local.yaml
The current set of parameters in the config are similar to the ones provided for the baseline in the Epic-Kitchens paper.
The model can be trained on the full training set using the script train_epic_kitchens.py
.
cd <path to repo>/Epic_Kitchens_Object_Detector_Detectron python train_epic_kitchens.py --config-file config/<config_file>.yaml --root-dir <path_to_Epic_Kitchens_rgb_frames> --ann-dir <path_to_Epic_Kitchens_annotation_files>
For a comprehensive list of arguments available, run:
python train_epic_kitchens.py --help
The script does not yet support custom training and validation.
The script infer_epic_kitchens.sh
is provided to run the inference on the test object frames of Epic Kitchens. To run the inference, simply run this script as follows:
cd <path to repo>/Epic_Kitchens_Object_Detector_Detectron python infer_epic_kitchens.py --config-file config/<config_file>.yaml --root-dir <path_to_Epic_Kitchens_rgb_frames> --ann-dir <path_to_Epic_Kitchens_annotation_files>
The script does not yet support running inference on a custom set of videos. However you can always modify the annotation files EPIC_test_s1_object_video_list.csv
and/or EPIC_test_s2_object_video_list.csv
to specify which videos you want to infer for.
The feature extraction module is also not ready yet.
The detections of the test frames are saved in the default submission format for the Object Detection Challenge:
|---<path to output> | |---seen.json | |---unseen.json
This project is licensed under the MIT License - see the LICENSE file for details.
Please note, Detectron2 is licensed under Apache2.0. Please respect the original licenses as well.
@INPROCEEDINGS{Damen2018EPICKITCHENS, title={Scaling Egocentric Vision: The EPIC-KITCHENS Dataset}, author={Damen, Dima and Doughty, Hazel and Farinella, Giovanni Maria and Fidler, Sanja and Furnari, Antonino and Kazakos, Evangelos and Moltisanti, Davide and Munro, Jonathan and Perrett, Toby and Price, Will and Wray, Michael}, booktitle={European Conference on Computer Vision (ECCV)}, year={2018} }
还没有评论,说两句吧!
热门资源
seetafaceJNI
项目介绍 基于中科院seetaface2进行封装的JAVA...
spark-corenlp
This package wraps Stanford CoreNLP annotators ...
Keras-ResNeXt
Keras ResNeXt Implementation of ResNeXt models...
capsnet-with-caps...
CapsNet with capsule-wise convolution Project ...
shih-styletransfer
shih-styletransfer Code from Style Transfer ...
智能在线
400-630-6780
聆听.建议反馈
E-mail: support@tusaishared.com