CV

EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks This paper created a huge mark in the field of model scaling and parametric optimization in terms of model architecture. It brings us a new scaling method called compund scaling, to scale the convolution network in all the three dimensions - Width, Depth and, resolution/channels. Along with this novel way of scaling it also brings us a new family of architecture created using Neural Architecture Search called the EfficentNet Family. ...

EfficientNetV2: Smaller Models and Faster Training This very recent paper (1-month-old at the time of writing this) introduces EfficientNetV2, a new family of convolutional networks that have faster training speed and better parameter efficiency. Based on top of the EfficientNet this paper pushes the boundary of model scaling and architecture search by further optimizing the network by using training aware Neural Architecture Search (NAS) and scaling. It jointly optimizes training speed and parameter efficiency to create the lightest best-performing models. ...

MLP-MIXER: An all MLP Architecture for Vision This is a very recent paper that challenges the need for complicated transformer-based models for huge datasets and questions the inductive biases presently in place for the present image recognition tasks. This paper argues that given a huge dataset (size 100M+), the performance of traditional CNN-based architectures or the new transformer-based architectures are only marginally better than a classic MLP based architecture, thus questioning the inductive biases of both CNNs and Transformers for images. ...

PICK: Processing Key Information Extraction from Documents using Improved Graph Learning-Convolutional Networks This paper talks KIE(Key Information Extraction), which is to extract texts of a number of key fields from given documents, and save the texts to structured documents. It proposes a solution which overcomes the existing problems fully and efficiently exploiting both textual and visual features of documents to get a richer semantic representation that is crucial for extraction. ...