Image

DetectoRS: Detecting Objects with Recursive Feature Pyramid and Switchable Atrous Convolution [code] (Jun 2020)
Object detection combining macro features (Recursive Feature Pyramid) with micro features (Switchable Atrous Convolution).

Image Segmentation Using Deep Learning: A Survey (Jan 2020)
Survey of image segmentation.

Searching for MobileNetV3 (May 2019)
Improved Image network for mobile.

Visualizing Deep Similarity Networks (Jan 2019)
Visualize the image regions responsible for pairwise similarity in an embedding network.

CurriculumNet: Weakly Supervised Learning from Large-Scale Web Images [code] (Aug 2018)
Train CNN with weakly labeled data from the internet. The training is performed using easy subsets (fewer noise data) and increasing complexity (more noise data). The split is done by density clustering algorithm.

Tracking Emerges by Colorizing Videos (Jul 2018)
Unsupervised learning of colorizing videos make tracking of objects emerge as the network needs to learn to track object to paint them.

AutoAugment: Learning Augmentation Policies from Data (May 2018)
Method to learn augmentation techniques from the dataset.

Focal Loss for Dense Object Detection [code] (Feb 2018)
Loss based on a modulation factor in the cross-entropy to minimize the problem of classification in very imbalanced dataset.

Deformable Convolutional Networks [code] (Mar 2017)
Convolutions with offsets to create deformable convolutions, the offset is a trainable vector.

Spatial Transformer Networks [code] (Jun 2015)
Removes spatial invariance from images by applying a learnable affine transformation followed by interpolation.

Is object localization for free? – Weakly-supervised learning with convolutional neural networks (Mar 2015)
They present a method to perform object detection without bounding boxes, using a conv layer instead of a softmax to perform discriminative classification in a convolutional way and define several discriminative bounding boxes