Deep Learning References

Convolutional neural networks (CNNs) trained by backpropagation

• Y. LeCun, B. Boser, J. S. Denker, D. Henderson, R. E. Howard, W. Hubbard, L. D. Jackel. Backpropagation Applied to Handwritten Zip Code Recognition. Neural Computation, 1(4):541-551, 1989. DOI: 10.1162/neco.1989.1.4.541

• First use of convolutional neural networks (CNNs) trained by backpropagation.

Faster R-CNN

• S. Ren, K. He, R. Girshick, J. Sun. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks, Procs. of the 28th International Conference on Neural Information Processing Systems (NIPS'15), 1:91-99, 2015. Conference paper, arXiv:1506.01497.
• IEEE Transactions on Pattern Analysis and Machine Intelligence, 39(6), 2017. DOI: 10.1109/TPAMI.2016.2577031.

Mask R-CNN

• K. He, G. Gkioxari, P. Dollár, R. Girshick. Mask R-CNN, arXiv:1703.06870, 2017.

Transformer network

• A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, I. Polosukhin. Attention is all you need, 31st Conference on Neural Information Processing Systems (NIPS 2017), 2017. arXiv:1706.03762.

Panoptic segmentation

• A. Kirillov, K. He, R. Girshick, C. Rother, P. Dollár. Panoptic Segmentation, The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2019, pp. 9404-9413. arXiv:1801.00868.

DETR network

• N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov and S. Zagoruyko. End-to-End Object Detection with Transformers. arXiv:2005.12872, 2020.
• DETR network: In object detection, replace anchor boxes + proposals + Non Maximum Suppression by a Transformer that regresses a fixed number of boxes and classifies the object inside each box. One important class is "empty", for empty boxes.
• Comparable results to Faster R-CNN, but fewer FLOPS and number of parameters.
• Extended with panoptic segmentation head for instance segmentation of object within each box.