Summary of Contrastive Learning For Character Detection in Ancient Greek Papyri, by Vedasri Nakka et al.
Contrastive Learning for Character Detection in Ancient Greek Papyri
by Vedasri Nakka, Andreas Fischer, Rolf Ingold, Lars Vogtlin
First submitted to arxiv on: 16 Sep 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The thesis explores the effectiveness of SimCLR, a contrastive learning technique, in Greek letter recognition. The study compares SimCLR’s performance with traditional baseline models using cross-entropy and triplet loss functions on the ICDAR dataset. Additionally, it examines the role of different data augmentation strategies in the SimCLR training process. The research trains three primary approaches: a baseline model using cross-entropy loss, a triplet embedding model with a classification layer, and a SimCLR pretrained model with a classification layer. Initially, models are trained using 93 augmentations on ResNet-18 and ResNet-50 networks. The top four augmentations are selected using a statistical t-test. Pretraining of SimCLR is conducted on the Alpub dataset, followed by fine-tuning on the ICDAR dataset. The study shows that SimCLR does not outperform the baselines in letter recognition tasks. The baseline model with cross-entropy loss demonstrates better performance than both SimCLR and the triplet loss model. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary SimCLR is a technique for training models. In this study, researchers used it to recognize Greek letters. They compared its performance to traditional methods on a dataset called ICDAR. They also looked at how different ways of changing the images (augmentations) helped or hurt SimCLR’s results. The researchers tried three main approaches: one that uses cross-entropy loss, one that uses triplet loss, and one that uses SimCLR. First, they trained all three methods using many different augmentations. Then, they picked the four best ones. They pre-trained SimCLR on a bigger dataset called Alpub, then fine-tuned it on ICDAR. The results show that traditional methods are better at recognizing Greek letters than SimCLR. |
Keywords
* Artificial intelligence * Classification * Cross entropy * Data augmentation * Embedding * Fine tuning * Pretraining * Resnet * Triplet loss
