English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Machine learning techniques have excelled in the automatic semantic analysisof images, reaching human-level performances on challenging benchmarks. Yet,the semantic analysis of videos remains challenging due to the significantlyhigher dimensionality of the input data, respectively, the significantly higherneed for annotated training examples. By studying the automatic recognition ofGerman sign language videos, we demonstrate that on the relatively scarcetraining data of 2.800 videos, modern deep learning architectures for videoanalysis (such as ResNeXt) along with transfer learning on large gesturerecognition tasks, can achieve about 75% character accuracy. Considering thatthis leaves us with a probability of under 25% that a 5 letter word is spelledcorrectly, spell-correction systems are crucial for producing readable outputs.The contribution of this paper is to propose a convolutional neural network forspell-correction that expects the softmax outputs of the character recognitionnetwork (instead of a misspelled word) as an input. We demonstrate that purelylearning on softmax inputs in combination with scarce training data yieldsoverfitting as the network learns the inputs by heart. In contrast, trainingthe network on several variants of the logits of the classification output i.e.scaling by a constant factor, adding of random noise, mixing of softmax andhardmax inputs or purely training on hardmax inputs, leads to bettergeneralization while benefitting from the significant information hidden inthese outputs (that have 98% top-5 accuracy), yielding a readable text despitethe comparably low character accuracy.

Exploiting the Logits: Joint Sign Language Recognition and Spell-Correction