Lucid Data Dreaming for Object Tracking

نویسندگان

  • Anna Khoreva
  • Rodrigo Benenson
  • Eddy Ilg
  • Thomas Brox
  • Bernt Schiele
چکیده

Convolutional networks reach top quality in pixel-level object tracking but require a large amount of training data (1k∼10k) to deliver such results. We propose a new training strategy which achieves state-of-the-art results across three evaluation datasets while using 20×∼100× less annotated data than competing methods. Instead of using large training sets hoping to generalize across domains, we generate in-domain training data using the provided annotation on the first frame of each video to synthesize (“lucid dream”1) plausible future video frames. In-domain per-video training data allows us to train high quality appearanceand motion-based models, as well as tune the post-processing stage. This approach allows to reach competitive results even when training from only a single annotated frame, without ImageNet pre-training. Our results indicate that using a larger training set is not automatically better, and that for the tracking task a smaller training set that is closer to the target domain is more effective. This changes the mindset regarding how many training samples and general “objectness” knowledge are required for the object tracking task.

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Applications of Lucid Dreaming in Sports

The following article has above all a practical orientation. The various possibilities for the application of lucid dreaming in sports training are presented and briefly illustrated. These theses are based on findings from experiments with experienced lucid dreamers who were instructed to carry out various routine and sport-related actions while lucid-dreaming, with the object of observing the ...

متن کامل

Lucid dreaming incidence: A quality effects meta-analysis of 50years of research.

We report a quality effects meta-analysis on studies from the period 1966-2016 measuring either (a) lucid dreaming prevalence (one or more lucid dreams in a lifetime); (b) frequent lucid dreaming (one or more lucid dreams in a month) or both. A quality effects meta-analysis allows for the minimisation of the influence of study methodological quality on overall model estimates. Following sensiti...

متن کامل

Neural Correlates of Lucid Dreaming and Comparisons with Phenomenological Aspects

Research on the neural correlates of lucid dreaming has recently gained more underlying data. By exploring seven studies that investigated the neural basis of lucid dreaming, this essay sought to examine which neural correlates are associated with lucid dreaming and how proposed neural correlates relate to phenomenological aspects. Dorsolateral prefrontal cortex (DLPFC) was judged as the region...

متن کامل

Neural correlates of dream lucidity obtained from contrasting lucid versus non-lucid REM sleep: a combined EEG/fMRI case study.

STUDY OBJECTIVES To investigate the neural correlates of lucid dreaming. DESIGN Parallel EEG/fMRI recordings of night sleep. SETTING Sleep laboratory and fMRI facilities. PARTICIPANTS Four experienced lucid dreamers. INTERVENTIONS N/A. MEASUREMENTS AND RESULTS Out of 4 participants, one subject had 2 episodes of verified lucid REM sleep of sufficient length to be analyzed by fMRI. Dur...

متن کامل

Lucid Dreaming – Dreams of Clarity

Although the concept of ‘lucid dreaming’ was fi rst coined by Frederik Willems van Eeden, a Dutch psychiatrist, and introduced at the meeting of The Society for Psychical Research held on 22 April 1913, the phenomenon of lucid dreaming was already known in earlier historical periods and its descriptions can be found in writings of Aristotle. However, modern lucid dreaming research was establish...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

عنوان ژورنال:
  • CoRR

دوره abs/1703.09554  شماره 

صفحات  -

تاریخ انتشار 2017