Datanest

datanest

Saša Milić is uniquely placed to compare oracle networks objectively. On top of being a software engineer and data scientist, Saša is also a co-founder of API3. While she no longer works full-time at any network, Saša has worked closely with various oracle projects in the last few years; during her presentation at the Blockchain Oracle Summit, Saša shared “a smattering of hot takes” about some of the more widely used oracle networks in the space today – including Chainlink, UMA, Tellor, Nest and BAND. In this article, we’ll guide you through her presentation, sharing key concepts. https://youtu.be/c0aOAkjzTRA?embedable=true Chainlink Oracle Review The most widely used oracle network today, has a reputation for providing consistently good data. According to DeFi Pulse (at the time of presentation), 15 of the top 20 DeFi protocols by Total Value Locked (TVL) use Chainlink, making the network a cornerstone of the DeFi Ecosystem. Chainlink Chainlink works through a decentralized network of nodes that collect data from reliable sources and aggregate the results on-chain. Chainlink refers to this as “t ” because each oracle network, node operator, and the data source is selected in a decentralized way. hree layers of decentralization According to Chainlink, this in the network. eliminates any single point of failure Chainlink, established in 2017, is well-known for consistently providing reliable price data. Chainlink currently caters to the larger part of the DeFi ecosystem, with a majority of the top DeFi projects using their oracle services. Some projects using Chainlink price feeds include , , and . Aave Compound Synthetix Out of all the oracle projects currently on the market, Chainlink seems to be investing the most in cryptography research. According to Saša, almost half of the team at are qualified cryptographers. Chainlink has also integrated two Cornell projects, and . Town Crier is a system that connects smart contracts to external data from trusted websites while also ensuring the confidentiality of the data. Chainlink Labs Town Crier DECO DECO uses zero-knowledge proofs to verify off-chain data while maintaining full data privacy. Chainlink’s whitepapers were also co-authored by well-known cryptographer and Cornell Tech professor Ari Juels who is now the Chief Scientist at Chainlink Labs. Since Saša’s talk at BOS22 in June, Chainlink has that staking will be live later this year. Staking will provide the network with more crypto-economic security, as stakers will initially be tasked with identifying and reporting node operators that fail to meet predetermined standards. For more information on LINK staking, read . announced this article UMA Oracle Review At , security is formalized in an economic guarantee around the cost of corrupting an oracle through their Data Verification Mechanism (DVM). The way these economic guarantee works is based on the idea that every oracle has a minimum price needed to become corrupted. In most cases, this is the price needed to take control of 51% of the voting tokens and is called the Cost of Corruption (CoC). UMA On the other side of the CoC is the potential (PfC). This is the maximum amount of money that can be gained from corrupting an oracle. Therefore, if CoC is kept above PfC, there would be no economic incentive for someone to corrupt the oracle. UMA is designed to measure the PfC of all the contracts in their system, then keep CoC above that number through a series of token buybacks triggered when necessary – more information . Profit from Corruption can be found here UMA also employs the concept of the “Optimistic Oracle” based on their . In essence, when a contract requests data – the “requestor” – anyone can propose an answer – the “proposer.” dispute resolution model The proposer needs to post a bond together with their data. Anyone else can verify that data and raise a dispute to UMA’s DVM if the data is incorrect – the disputer. If no one disputes that data within a set amount of time, the proposer’s bond is released back to them, and they are paid a pre-agreed fee. Because disputes are economically incentivized, a lack of disputes can be seen as verification that a piece of data is true. This mechanism is what makes UMA an ‘optimistic’ oracle. Tellor Oracle Review was one of the earlier projects to talk about (OEV). Simply put, OEV is the potential profit a party could earn by being in the position to update the oracle. Similar in principle to Miner Extractable Value (MEV), if only one party is allowed to update an oracle, they would be in a unique position to take advantage of the difference in prices before they make the update. Tellor Oracle Extractable Value For example, if party A has control of the ETH/USD price feed updates on a certain exchange, then that party can make trades before updating the prices since they know which way the price is going. Tellor approached this potential problem by building an oracle protocol that is truly decentralized, allowing anyone to be a data provider. Their protocol takes advantage of OEV as a means of making payments to oracle data providers rather than using gas fees. A more in-depth explanation of OEV can be found in written by Nicholas Fett, CTO of Tellor. this article Tellor has also been integrated as a fallback oracle in the Liquity protocol’s , which is significant given the effort that the team at Liquity has put into oracle research and their top position on the list. dual-oracle system DeFi Safety Conversely, Tellor’s dispute resolution system has a natural paradox. The system works by paying people to dispute incorrect data. If they raise their dispute successfully, they are paid in Tribute (TRB), Tellor’s native token. The presence of disputers, therefore, works to discourage false reports in the first place. However, since disputers profit from reporting faulty data, discouraging false reports would decrease the demand for disputers since no faulty reports mean no one can profit from disputing. This, in turn, leads to no one watching the data to dispute its accuracy, which then creates more opportunities for false reports. This, as Saša notes, is not a problem unique to Tellor but is common in optimistic oracle systems. Nest Oracle Review is currently only used for price feeds, and miners provide the data for these prices. Nest When miners report a price, they have to put up the same amount of money that they are reporting. This means that if they indicate the wrong price, it creates an opportunity for someone to take advantage of the difference between the reported and actual prices to make a profit. Band “Cross-chain data oracle.” According to Saša’s research, has relatively high adoption within DeFi projects. Despite that, there has been only one known exploit of Band, indicating their overall reliability. Band Protocol Band’s are speed and scalability – to process a large number of data requests within seconds; cross-chain compatibility – to be able to serve most of the available blockchains, and data flexibility – to handle different types of data and the different ways they are retrieved. This, according to Saša, is lacking a key area – security. design goals Band also runs a separate blockchain called . This makes it harder to evaluate its security because the security of the oracle is dependent on the overall security of this sidechain – explained here by . Bandchain CoinDesk Additionally, BAND holders do not participate in validation directly on Bandchain. Instead, they delegate their tokens to validators, who run validation nodes and retrieve data from the data sources. In each request, validators are randomly selected to retrieve and verify the data before committing them to the blockchain – more information about their system can be found . Given the vital role that validators play in this system, Saša  is unsure if the low number of validators (67 actives at the time of publication, based on ) here Cosmoscan The Blockchain Oracle Summit was the world’s first conference to focus solely on the importance of oracles and their design. Leading builders and researchers from around the world gathered in Berlin to share their work and discuss challenges faced by a close-knit group of hardcore oracle enthusiasts. To keep up with more of Saša’s hot takes, follow her on . Twitter

Saša Milić Offers Hot-Takes on Chainlink, UMA, Tellor, Nest, and Band Protocol

How K-SIF and SIF Revolutionize Anomaly Detection in Complex Datasets

Authors:
(1) Nora Schneider, Computer Science Department, ETH Zurich, Zurich, Switzerland (nschneide@student.ethz.ch);
(2) Shirin Goshtasbpour, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (shirin.goshtasbpour@inf.ethz.ch);
(3) Fernando Perez-Cruz, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (fernando.perezcruz@sdsc.ethz.ch). Table of Links Abstract and 1 Introduction 2 Background 2.1 Data Augmentation 2.2 Anchor Regression 3 Anchor Data Augmentation 3.1 Comparison to C-Mixup and 3.2 Preserving nonlinear data structure 3.3 Algorithm 4 Experiments and 4.1 Linear synthetic data 4.2 Housing nonlinear regression 4.3 In-distribution Generalization 4.4 Out-of-distribution Robustness 5 Conclusion, Broader Impact, and References A Additional information for Anchor Data Augmentation B Experiments 4.2 Housing nonlinear regression We extend the results from the previous section to the California and Boston housing data and compare ADA to C-Mixup [49]. We repeat the same experiments on three different regression datasets. Results are provided in Appendix B.2 and also show the superiority of ADA over C-Mixup for data augmentation in the implemented experimental setup. Data: We use the California housing dataset [19] and the Boston housing dataset [14]. The training dataset contains up to n = 406 samples, and the remaining samples are for validation. We report the results as a function of the number of training points. Models and comparisons: We fit a ridge regression model (baseline) and train a MLP with one hidden layer with a varying number of hidden units with sigmoid activation. The baseline models only use only the original data. We train the same models using C-Mixup with a Gaussian kernel and bandwidth of 1.75. We compare the previous approaches to models fitted on ADA augmented data. We generate 20 different augmentations per original observation using different values for γ controlled via α = 4 similar to what was described in Section 4.1. The Anchor matrix is constructed using k-means clustering with q = 10. Results: We report the results in Figure 3. First, we observe that the MLPs outperform Ridge regression suggesting a nonlinear data structure. Second, when the number of training samples is low, applying ADA improves the performance of all models compared to C-Mixup and the baseline. The performance gap decreases as the number of samples increases. When comparing C-Mixup and ADA, we see that using sufficiently many samples both methods achieve similar performance. While on the Boston data, the performance gap between the baseline and ADA persists, on California housing, the non-augmented model fit performs better than the augmented one when data availability increases. This suggests that there is a sweet spot where the addition of original data samples is required for better generalization, and augmented samples cannot contribute any further. This paper is available on arxiv under CC0 1.0 DEED license. Authors: (1) Nora Schneider, Computer Science Department, ETH Zurich, Zurich, Switzerland (nschneide@student.ethz.ch); (2) Shirin Goshtasbpour, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (shirin.goshtasbpour@inf.ethz.ch); (3) Fernando Perez-Cruz, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (fernando.perezcruz@sdsc.ethz.ch). Authors: Authors: (1) Nora Schneider, Computer Science Department, ETH Zurich, Zurich, Switzerland (nschneide@student.ethz.ch); (2) Shirin Goshtasbpour, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (shirin.goshtasbpour@inf.ethz.ch); (3) Fernando Perez-Cruz, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (fernando.perezcruz@sdsc.ethz.ch). Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Background 2.1 Data Augmentation 2.1 Data Augmentation 2.2 Anchor Regression 2.2 Anchor Regression 3 Anchor Data Augmentation 3 Anchor Data Augmentation 3.1 Comparison to C-Mixup and 3.2 Preserving nonlinear data structure 3.1 Comparison to C-Mixup and 3.2 Preserving nonlinear data structure 3.3 Algorithm 3.3 Algorithm 4 Experiments and 4.1 Linear synthetic data 4 Experiments and 4.1 Linear synthetic data 4.2 Housing nonlinear regression 4.2 Housing nonlinear regression 4.3 In-distribution Generalization 4.3 In-distribution Generalization 4.4 Out-of-distribution Robustness 4.4 Out-of-distribution Robustness 5 Conclusion, Broader Impact, and References 5 Conclusion, Broader Impact, and References A Additional information for Anchor Data Augmentation A Additional information for Anchor Data Augmentation B Experiments B Experiments 4.2 Housing nonlinear regression We extend the results from the previous section to the California and Boston housing data and compare ADA to C-Mixup [49]. We repeat the same experiments on three different regression datasets. Results are provided in Appendix B.2 and also show the superiority of ADA over C-Mixup for data augmentation in the implemented experimental setup. Data : We use the California housing dataset [19] and the Boston housing dataset [14]. The training dataset contains up to n = 406 samples, and the remaining samples are for validation. We report the results as a function of the number of training points. Data Models and comparisons: We fit a ridge regression model (baseline) and train a MLP with one hidden layer with a varying number of hidden units with sigmoid activation. The baseline models only use only the original data. We train the same models using C-Mixup with a Gaussian kernel and bandwidth of 1.75. We compare the previous approaches to models fitted on ADA augmented data. We generate 20 different augmentations per original observation using different values for γ controlled via α = 4 similar to what was described in Section 4.1. The Anchor matrix is constructed using k-means clustering with q = 10. Models and comparisons: Results : We report the results in Figure 3. First, we observe that the MLPs outperform Ridge regression suggesting a nonlinear data structure. Second, when the number of training samples is low, applying ADA improves the performance of all models compared to C-Mixup and the baseline. The performance gap decreases as the number of samples increases. When comparing C-Mixup and ADA, we see that using sufficiently many samples both methods achieve similar performance. While on the Boston data, the performance gap between the baseline and ADA persists, on California housing, the non-augmented model fit performs better than the augmented one when data availability increases. This suggests that there is a sweet spot where the addition of original data samples is required for better generalization, and augmented samples cannot contribute any further. Results This paper is available on arxiv under CC0 1.0 DEED license. This paper is available on arxiv under CC0 1.0 DEED license. available on arxiv

ADA vs C-Mixup: Performance on California and Boston Housing Datasets

Authors:
(1) Nora Schneider, Computer Science Department, ETH Zurich, Zurich, Switzerland (nschneide@student.ethz.ch);
(2) Shirin Goshtasbpour, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (shirin.goshtasbpour@inf.ethz.ch);
(3) Fernando Perez-Cruz, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (fernando.perezcruz@sdsc.ethz.ch). Table of Links Abstract and 1 Introduction 2 Background 2.1 Data Augmentation 2.2 Anchor Regression 3 Anchor Data Augmentation 3.1 Comparison to C-Mixup and 3.2 Preserving nonlinear data structure 3.3 Algorithm 4 Experiments and 4.1 Linear synthetic data 4.2 Housing nonlinear regression 4.3 In-distribution Generalization 4.4 Out-of-distribution Robustness 5 Conclusion, Broader Impact, and References A Additional information for Anchor Data Augmentation B Experiments 4 Experiments We experimentally investigate and compare the performance of ADA. First, we use ADA in an in-distribution setting for a linear regression problem (Section 4.1), in which we show that even in this case, ADA provides improved performance in the low data regime. Second, in Section 4.2, we apply ADA and C-Mixup to the California and Boston Housing datasets as we increase the number of training samples. In the last two subsections, we replicate the in-distribution generalization (Section 4.3) and the out-of-distribution Robustness (Section 4.4) from the C-Mixup paper [49]. In [49] the authors further assess a task generalization experiment. However, the corresponding code was not publicly provided, and a comparison could not be easily made. 4.1 Linear synthetic data Using synthetic linear data, we investigate if ADA can improve model performance in an overparameterized setting compared to C-Mixup, vanilla augmentation, or classical expected risk minimization (ERM). Additionally, we analyze the sensitivity of our approach to the choice of γ and the number of augmentations. Data: The generated data follows a standard linear structure Models and Comparisons: We investigate and compare the impact of ADA using two different models with varying complexity: a linear Ridge regression and a multilayer perceptron (MLP) with one hidden layer and 10 units with ReLU activation. Using an MLP with more hidden layers shows similar results (see Appendix B.1 for details). For the Ridge regression model, we increase the dataset by a factor of 10 by sampling from the respective augmentation methods and subsequently compute the regression estimators. In contrast, for the MLP, we implement the augmentation methods on a minibatch level. Specifically, we incorporate vanilla augmentation by adding Gaussian noise to each batch, apply C-Mixup after sampling from the beta distribution in each batch, and finally, apply ADA after sampling from the defined gamma values in each batch. In summary, even in the simplest of cases, in which we should not expect gains from ADA (or C-Mixup), these data augmentation strategies provide gains in performance when the number of training examples is not sufficient to achieve the error floor. This paper is available on arxiv under CC0 1.0 DEED license. Authors: (1) Nora Schneider, Computer Science Department, ETH Zurich, Zurich, Switzerland (nschneide@student.ethz.ch); (2) Shirin Goshtasbpour, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (shirin.goshtasbpour@inf.ethz.ch); (3) Fernando Perez-Cruz, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (fernando.perezcruz@sdsc.ethz.ch). Authors: Authors: (1) Nora Schneider, Computer Science Department, ETH Zurich, Zurich, Switzerland (nschneide@student.ethz.ch); (2) Shirin Goshtasbpour, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (shirin.goshtasbpour@inf.ethz.ch); (3) Fernando Perez-Cruz, Computer Science Department, ETH Zurich, Zurich, Switzerland and Swiss Data Science Center, Zurich, Switzerland (fernando.perezcruz@sdsc.ethz.ch). Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Background 2.1 Data Augmentation 2.1 Data Augmentation 2.2 Anchor Regression 2.2 Anchor Regression 3 Anchor Data Augmentation 3 Anchor Data Augmentation 3.1 Comparison to C-Mixup and 3.2 Preserving nonlinear data structure 3.1 Comparison to C-Mixup and 3.2 Preserving nonlinear data structure 3.3 Algorithm 3.3 Algorithm 4 Experiments and 4.1 Linear synthetic data 4 Experiments and 4.1 Linear synthetic data 4.2 Housing nonlinear regression 4.2 Housing nonlinear regression 4.3 In-distribution Generalization 4.3 In-distribution Generalization 4.4 Out-of-distribution Robustness 4.4 Out-of-distribution Robustness 5 Conclusion, Broader Impact, and References 5 Conclusion, Broader Impact, and References A Additional information for Anchor Data Augmentation A Additional information for Anchor Data Augmentation B Experiments B Experiments 4 Experiments We experimentally investigate and compare the performance of ADA. First, we use ADA in an in-distribution setting for a linear regression problem (Section 4.1), in which we show that even in this case, ADA provides improved performance in the low data regime. Second, in Section 4.2, we apply ADA and C-Mixup to the California and Boston Housing datasets as we increase the number of training samples. In the last two subsections, we replicate the in-distribution generalization (Section 4.3) and the out-of-distribution Robustness (Section 4.4) from the C-Mixup paper [49]. In [49] the authors further assess a task generalization experiment. However, the corresponding code was not publicly provided, and a comparison could not be easily made. 4.1 Linear synthetic data Using synthetic linear data, we investigate if ADA can improve model performance in an overparameterized setting compared to C-Mixup, vanilla augmentation, or classical expected risk minimization (ERM). Additionally, we analyze the sensitivity of our approach to the choice of γ and the number of augmentations. Data : The generated data follows a standard linear structure Data Models and Comparisons: We investigate and compare the impact of ADA using two different models with varying complexity: a linear Ridge regression and a multilayer perceptron (MLP) with one hidden layer and 10 units with ReLU activation. Using an MLP with more hidden layers shows similar results (see Appendix B.1 for details). Models and Comparisons: For the Ridge regression model, we increase the dataset by a factor of 10 by sampling from the respective augmentation methods and subsequently compute the regression estimators. In contrast, for the MLP, we implement the augmentation methods on a minibatch level. Specifically, we incorporate vanilla augmentation by adding Gaussian noise to each batch, apply C-Mixup after sampling from the beta distribution in each batch, and finally, apply ADA after sampling from the defined gamma values in each batch. In summary, even in the simplest of cases, in which we should not expect gains from ADA (or C-Mixup), these data augmentation strategies provide gains in performance when the number of training examples is not sufficient to achieve the error floor. This paper is available on arxiv under CC0 1.0 DEED license. This paper is available on arxiv under CC0 1.0 DEED license. available on arxiv

Evaluating ADA: Experimental Results on Linear and Housing Datasets

Authors:
(1) Zhaoqing Wang, The University of Sydney and AI2Robotics;
(2) Xiaobo Xia, The University of Sydney;
(3) Ziye Chen, The University of Melbourne;
(4) Xiao He, AI2Robotics;
(5) Yandong Guo, AI2Robotics;
(6) Mingming Gong, The University of Melbourne and Mohamed bin Zayed University of Artificial Intelligence;
(7) Tongliang Liu, The University of Sydney. Table of Links Abstract and 1. Introduction 2. Related works 3. Method and 3.1. Problem definition 3.2. Baseline and 3.3. Uni-OVSeg framework 4. Experiments 4.1. Implementation details 4.2. Main results 4.3. Ablation study 5. Conclusion 6. Broader impacts and References A. Framework details B. Promptable segmentation C. Visualisation C. Visualisation We illustrate a wide range of visualisations of promptable segmentation and open-vocabulary segmentation across multiple datasets. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Zhaoqing Wang, The University of Sydney and AI2Robotics; (2) Xiaobo Xia, The University of Sydney; (3) Ziye Chen, The University of Melbourne; (4) Xiao He, AI2Robotics; (5) Yandong Guo, AI2Robotics; (6) Mingming Gong, The University of Melbourne and Mohamed bin Zayed University of Artificial Intelligence; (7) Tongliang Liu, The University of Sydney. Authors: Authors: (1) Zhaoqing Wang, The University of Sydney and AI2Robotics; (2) Xiaobo Xia, The University of Sydney; (3) Ziye Chen, The University of Melbourne; (4) Xiao He, AI2Robotics; (5) Yandong Guo, AI2Robotics; (6) Mingming Gong, The University of Melbourne and Mohamed bin Zayed University of Artificial Intelligence; (7) Tongliang Liu, The University of Sydney. Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2. Related works 2. Related works 3. Method and 3.1. Problem definition 3. Method and 3.1. Problem definition 3.2. Baseline and 3.3. Uni-OVSeg framework 3.2. Baseline and 3.3. Uni-OVSeg framework 4. Experiments 4.1. Implementation details 4.1. Implementation details 4.2. Main results 4.2. Main results 4.3. Ablation study 4.3. Ablation study 5. Conclusion 5. Conclusion 6. Broader impacts and References 6. Broader impacts and References A. Framework details A. Framework details B. Promptable segmentation B. Promptable segmentation C. Visualisation C. Visualisation C. Visualisation We illustrate a wide range of visualisations of promptable segmentation and open-vocabulary segmentation across multiple datasets. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

Visualizing Promptable and Open-Vocabulary Segmentation Across Multiple Datasets

Authors:
(1) Zhaoqing Wang, The University of Sydney and AI2Robotics;
(2) Xiaobo Xia, The University of Sydney;
(3) Ziye Chen, The University of Melbourne;
(4) Xiao He, AI2Robotics;
(5) Yandong Guo, AI2Robotics;
(6) Mingming Gong, The University of Melbourne and Mohamed bin Zayed University of Artificial Intelligence;
(7) Tongliang Liu, The University of Sydney. Table of Links Abstract and 1. Introduction 2. Related works 3. Method and 3.1. Problem definition 3.2. Baseline and 3.3. Uni-OVSeg framework 4. Experiments 4.1. Implementation details 4.2. Main results 4.3. Ablation study 5. Conclusion 6. Broader impacts and References A. Framework details B. Promptable segmentation C. Visualisation B. Promptable segmentation. Evaluation details. We perform a prompt segmentation evaluation on a wide range of datasets in various domains. For the point prompt, we adopt a uniform point grid h×w as input prompts (e.g., 20 × 20). For the box prompt, we use ground-truth bounding boxes as input prompts. 1-pt IoU denotes the oracle performance of one point by evaluating the intersection-overunion (IoU) of the predicted masks that best match ground truth. 1-box IoU denotes is similar to 1-pt IoU. More evaluation results are reported in Fig. 7, Fig. 8 and Fig. 9. Dataset details. A description of each dataset is given in Tab. 5. The iShape dataset has 6 subsets, including antenna, branch, fence, hanger, log and wire. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Zhaoqing Wang, The University of Sydney and AI2Robotics; (2) Xiaobo Xia, The University of Sydney; (3) Ziye Chen, The University of Melbourne; (4) Xiao He, AI2Robotics; (5) Yandong Guo, AI2Robotics; (6) Mingming Gong, The University of Melbourne and Mohamed bin Zayed University of Artificial Intelligence; (7) Tongliang Liu, The University of Sydney. Authors: Authors: (1) Zhaoqing Wang, The University of Sydney and AI2Robotics; (2) Xiaobo Xia, The University of Sydney; (3) Ziye Chen, The University of Melbourne; (4) Xiao He, AI2Robotics; (5) Yandong Guo, AI2Robotics; (6) Mingming Gong, The University of Melbourne and Mohamed bin Zayed University of Artificial Intelligence; (7) Tongliang Liu, The University of Sydney. Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2. Related works 2. Related works 3. Method and 3.1. Problem definition 3. Method and 3.1. Problem definition 3.2. Baseline and 3.3. Uni-OVSeg framework 3.2. Baseline and 3.3. Uni-OVSeg framework 4. Experiments 4.1. Implementation details 4.1. Implementation details 4.2. Main results 4.2. Main results 4.3. Ablation study 4.3. Ablation study 5. Conclusion 5. Conclusion 6. Broader impacts and References 6. Broader impacts and References A. Framework details A. Framework details B. Promptable segmentation B. Promptable segmentation C. Visualisation C. Visualisation B. Promptable segmentation. Evaluation details. We perform a prompt segmentation evaluation on a wide range of datasets in various domains. For the point prompt, we adopt a uniform point grid h×w as input prompts (e.g., 20 × 20). For the box prompt, we use ground-truth bounding boxes as input prompts. 1-pt IoU denotes the oracle performance of one point by evaluating the intersection-overunion (IoU) of the predicted masks that best match ground truth. 1-box IoU denotes is similar to 1-pt IoU. More evaluation results are reported in Fig. 7, Fig. 8 and Fig. 9. Evaluation details. Dataset details. A description of each dataset is given in Tab. 5. The iShape dataset has 6 subsets, including antenna, branch, fence, hanger, log and wire. Dataset details. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

Evaluating Promptable Segmentation with Uniform Point Grids and Bounding Boxes on Diverse Datasets

Summarizing Large Datasets of Customer Feedback Using Retrieval-Augmented Generation (RAG)

Authors:
(1) Zhaoqing Wang, The University of Sydney and AI2Robotics;
(2) Xiaobo Xia, The University of Sydney;
(3) Ziye Chen, The University of Melbourne;
(4) Xiao He, AI2Robotics;
(5) Yandong Guo, AI2Robotics;
(6) Mingming Gong, The University of Melbourne and Mohamed bin Zayed University of Artificial Intelligence;
(7) Tongliang Liu, The University of Sydney. Table of Links Abstract and 1. Introduction 2. Related works 3. Method and 3.1. Problem definition 3.2. Baseline and 3.3. Uni-OVSeg framework 4. Experiments 4.1. Implementation details 4.2. Main results 4.3. Ablation study 5. Conclusion 6. Broader impacts and References A. Framework details B. Promptable segmentation C. Visualisation 4.1. Implementation details Datasets. During training, we randomly sample the 30% subset from the SA-1B dataset [34], which contains ∼ 3 million images and ∼ 0.3 billion masks. Although this supervision provides diverse binary masks, it lacks the semantic class for each mask. In addition, following Chen et al. [8], we collect about 1.3 million image-text pairs and use a large vision-language model to refine them. Afterward, we use the ChatGPT-based parser to extract entities with descriptive words from these text descriptions. Evaluation & metrics. We evaluate our model mainly on three tasks, including open-vocabulary semantic segmentation, open-vocabulary panoptic segmentation, and promptable segmentation. Following previous work [76], we adopt prompt engineering from [21, 66] and prompt templates from [22, 37]. For open-vocabulary semantic segmentation, we zero-shot evaluate the model on the COCO [40], ADE20K [84], PASCAL [18] datasets. The open-vocabulary semantic segmentation results are evaluated with the mean Intersection-over-Union (mIoU). For open-vocabulary panoptic segmentation, we evaluate the model on the COCO, ADE20K, and Cityscapes [15] datasets. We report the panoptic quality (PQ), semantic quality (SQ), and recognition quality (RQ) for open-vocabulary panoptic segmentation. For promptable segmentation, we report the 1-Point and 1-Box IoU (Oracle) on a wide range of datasets. Oracle denotes that we select the output mask with the max IoU by calculating the IoU between the prediction and target mask. More details can be found in Appendix B. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Zhaoqing Wang, The University of Sydney and AI2Robotics; (2) Xiaobo Xia, The University of Sydney; (3) Ziye Chen, The University of Melbourne; (4) Xiao He, AI2Robotics; (5) Yandong Guo, AI2Robotics; (6) Mingming Gong, The University of Melbourne and Mohamed bin Zayed University of Artificial Intelligence; (7) Tongliang Liu, The University of Sydney. Authors: Authors: (1) Zhaoqing Wang, The University of Sydney and AI2Robotics; (2) Xiaobo Xia, The University of Sydney; (3) Ziye Chen, The University of Melbourne; (4) Xiao He, AI2Robotics; (5) Yandong Guo, AI2Robotics; (6) Mingming Gong, The University of Melbourne and Mohamed bin Zayed University of Artificial Intelligence; (7) Tongliang Liu, The University of Sydney. Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2. Related works 2. Related works 3. Method and 3.1. Problem definition 3. Method and 3.1. Problem definition 3.2. Baseline and 3.3. Uni-OVSeg framework 3.2. Baseline and 3.3. Uni-OVSeg framework 4. Experiments 4.1. Implementation details 4.1. Implementation details 4.2. Main results 4.2. Main results 4.3. Ablation study 4.3. Ablation study 5. Conclusion 5. Conclusion 6. Broader impacts and References 6. Broader impacts and References A. Framework details A. Framework details B. Promptable segmentation B. Promptable segmentation C. Visualisation C. Visualisation 4.1. Implementation details Datasets . During training, we randomly sample the 30% subset from the SA-1B dataset [34], which contains ∼ 3 million images and ∼ 0.3 billion masks. Although this supervision provides diverse binary masks, it lacks the semantic class for each mask. In addition, following Chen et al. [8], we collect about 1.3 million image-text pairs and use a large vision-language model to refine them. Afterward, we use the ChatGPT-based parser to extract entities with descriptive words from these text descriptions. Datasets Evaluation & metrics. We evaluate our model mainly on three tasks, including open-vocabulary semantic segmentation, open-vocabulary panoptic segmentation, and promptable segmentation. Following previous work [76], we adopt prompt engineering from [21, 66] and prompt templates from [22, 37]. For open-vocabulary semantic segmentation, we zero-shot evaluate the model on the COCO [40], ADE20K [84], PASCAL [18] datasets. The open-vocabulary semantic segmentation results are evaluated with the mean Intersection-over-Union (mIoU). For open-vocabulary panoptic segmentation, we evaluate the model on the COCO, ADE20K, and Cityscapes [15] datasets. We report the panoptic quality (PQ), semantic quality (SQ), and recognition quality (RQ) for open-vocabulary panoptic segmentation. For promptable segmentation, we report the 1-Point and 1-Box IoU (Oracle) on a wide range of datasets. Oracle denotes that we select the output mask with the max IoU by calculating the IoU between the prediction and target mask. More details can be found in Appendix B. Evaluation & metrics. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

Stories, News & Mentions about DATANEST

Datanest’s Stories on HackerNoon