Skip to content
프로젝트

bertinetto/cfnet has 523 GitHub stars and 150 forks. GitHot measured 0 stars in 24 hours and 0 in 7 days. Last checked 2026-11-21 08:13 UTC. Last push 2018-05-31.

이 수치를 측정하는 방법

측정된 스타 증가

in 24h
0
in 7 days
0
in 15 days
기록 수집 중
in 1 month
기록 수집 중

서로 겹치는 기간의 순증가량이며 일별 합계가 아닙니다. 측정값이 없으면 비워 둡니다. 스타는 관심도를 나타내며 품질을 보장하지 않습니다.

소개

[CVPR'17] Training a Correlation Filter end-to-end allows lightweight networks of 2 layers (600 kB) to high performance at fast speed..

머신러닝

README.md

Original

End-to-end representation learning for Correlation Filter based tracking

pipeline image


Project page: [http://www.robots.ox.ac.uk/~luca/cfnet.html]


WARNING: we used Matlab 2015, MatConvNet v1.0beta24, CUDA 8.0 and cudnn 5.1. Other configurations might work, but it is not guaranteed. In particular, we received several reports of problems with Matlab 2017.


Getting started

[ Tracking only ] If you don't care about training, you can simply use one of our pretrained networks with our basic tracker.

  1. Prerequisites: GPU, CUDA (we used 7.5), cuDNN (we used v5.1), Matlab, MatConvNet.
  2. Clone the repository.
  3. Download the pretrained networks from here and unzip the archive in cfnet/pretrained.
  4. Go to cfnet/src/tracking/ and remove the trailing .example from env_paths_tracking.m.example, startup.m.example, editing the files as appropriate.
  5. Be sure to have at least one video sequence in the appropriate format. The easiest thing to do is to download the validation set (from here) that we used for the tracking evaluation and then extract the validation folder in cfnet/data/.
  6. Start from one of the cfnet/src/tracking/run_*_evaluation.m entry points.

[ Training and tracking ] Start here if instead you prefer to DIY and train your own networks.

  1. Prerequisites: GPU, CUDA (we used 7.5), cuDNN (we used v5.1), Matlab, MatConvNet.
  2. Clone the repository.
  3. Follow these step-by-step instructions, which will help you generating a curated dataset compatible with the rest of the code.
  4. If you did not generate your own metadata, download imdb_video_2016-10.mat (6.7GB) with all the metadata and also the dataset stats. Put them in cfnet/data/.
  5. Go to cfnet/src/training and remove the trailing .example from env_paths_training.m.example and startup.m.example, editing the files as appropriate.
  6. The various cfnet/train/run_experiment_*.m are some examples to start training. Default hyper-params are at the start of experiment.m and are overwritten by custom ones specified in run_experiment_*.m.
  7. By default, training plots are saved in cfnet/src/training/data/. When you are happy, grab a network snapshot (net-epoch-X.mat) and save it somewhere (e.g. cfnet/pretrained/).
  8. Go to point 4. of Tracking only, follow the instructions and enjoy the labour of your own GPUs!