Datasets › InDL

InDL (In-Diagram Logic)

Introduced by Haobo Yang et al. in InDL: A New Dataset and Benchmark for In-Diagram Logic Interpretation based on Visual Illusion28 May 2023 archive 2025-07-28

Dataset Introduction

In this work, we introduce the In-Diagram Logic (InDL) dataset, an innovative resource crafted to rigorously evaluate the logic interpretation abilities of deep learning models. This dataset leverages the complex domain of visual illusions, providing a unique challenge for these models.

The InDL dataset is characterized by its intricate assembly of optical illusions, wherein each instance poses a specific logic interpretation challenge. These illusions are constructed based on six classic geometric optical illusions, known for their intriguing interplay between perception and logic.

Motivations and Content

The motivation behind the creation of the InDL dataset arises from a recognized gap in current deep learning research. While models have exhibited remarkable proficiency in various domains such as image recognition and natural language processing, their performance in tasks requiring logical reasoning remains less understood and often opaque due to their inherent 'black box' characteristics. By using the medium of visual illusions, the InDL dataset aims to probe these models in a unique and challenging way, helping to illuminate their logic interpretation capabilities.

The InDL dataset is a comprehensive collection of instances where each visual illusion varies in illusion strength. The strength signifies the degree of distortion introduced to challenge the models' logic interpretation. Hence, the dataset not only offers a complexity gradient for model evaluation but also allows the analysis of model performance against varying degrees of challenge intensity.

Potential Use Cases

The potential use cases of the InDL dataset are extensive. Beyond the primary goal of evaluating deep learning models' logic interpretation abilities, it also presents a robust tool for researchers to investigate how models react to visual perception challenges. This opens avenues to understand how these models can be improved and how their decision-making processes can be better interpreted.

Additionally, the InDL dataset could provide a rich testing ground for model developers. Its diverse and challenging instances could allow them to rigorously benchmark their models and detect potential weaknesses that might be overlooked in more conventional datasets.

Furthermore, the InDL dataset could serve as a valuable resource for teaching and learning purposes. It provides a visually engaging and intellectually stimulating way to explore the capabilities and limitations of deep learning models, particularly in the realm of logic interpretation.

Benchmarks archive 2025-07-28

All 1 leaderboard whose dataset resolves to this page shown (sort by any header). "First row" is the archive's own first row at snapshot, in the archive's row order; nothing here re-ranks and metric direction is not asserted.

First row (archive order)PaperCode
Classification InDL ConvNext Average Recall 93.47% A ConvNet for the 2020s keras-team/keras +53 9 Compare

Papers archive 2025-07-28

papers with a benchmark row: 9 · with a code link: 9 · where Syntology ran a sample: 7 (6 with a run with no instrument failure, 1 where every run was a failure of Syntology's instrument) Syntology

Show: all papers with a benchmark rowonly where code ran (7 of 9 with a benchmark row: 6 with a run with no instrument failure, 1 where every run was a failure of Syntology's instrument)

Page 1 of 1: papers 1 to 9 of 9 with a leaderboard row on this dataset's benchmarks, newest first by the archive's date (ties by slug; undated papers last).

The archive's own "papers using this dataset" list was never published, so this is the benchmark-backed subset, not that list; the archive's count for this dataset is 10. The Syntology column is from Syntology's graph, stated per sample; it is not part of any archive number. A line reads “N ran (of which C constructed an object rather than computing a result; K with no instrument failure: H honoured, V violated, P with no contract checked; I where Syntology's instrument failed) · U unverified” (C is a part of N, never taken away from it); the figure “where Syntology's instrument failed” counts failures of Syntology's instrument, not of the code. “Ran” means the sample executed on a synthesized fixture, not that the paper's result was reproduced. When the archive marks a repository official for the paper, the cell starts with that repository's state (the archive's flag, not a verdict on who wrote the code; “community repositories only” when every sample that ran came from a community repository, “official: no sample here; runs from other or unrecorded repositories” when some came from a repository the paper names or has in its text, or from none recorded); hover it for the repositories the samples that ran came from. Syntology's record for this page last changed ; when this build read Syntology's graph is in the build record.

PaperCodeResultsDateSamples run Syntology
A ConvNet for the 2020s 54 1 10 Jan 2022 community repositories only · 57 ran (of which 39 constructed an object rather than computing a result; 49 with no instrument failure: 2 honoured, 0 violated, 47 with no contract checked; 8 where Syntology's instrument failed) · 23 unverified (12 pointer-only for licence)
ResNet strikes back: An improved training procedure in timm 14 1 1 Oct 2021 official: harvested for another paper · 0 ran · 3 unverified
Searching for MobileNetV3 67 1 6 May 2019 86 ran (of which 22 constructed an object rather than computing a result; 75 with no instrument failure: 9 honoured, 2 violated, 64 with no contract checked; 11 where Syntology's instrument failed) · 19 unverified (46 pointer-only for licence)
YOLOv3: An Incremental Improvement 311 1 8 Apr 2018 94 ran (of which 0 constructed an object rather than computing a result; 82 with no instrument failure: 7 honoured, 2 violated, 73 with no contract checked; 12 where Syntology's instrument failed) · 30 unverified (24 pointer-only for licence)
Learning Transferable Architectures for Scalable Image Recognition 17 1 21 Jul 2017 0 ran · 1 unverified (1 pointer-only for licence)
Xception: Deep Learning with Depthwise Separable Convolutions 41 1 7 Oct 2016 7 ran (of which 0 constructed an object rather than computing a result; 6 with no instrument failure: 0 honoured, 0 violated, 6 with no contract checked; 1 where Syntology's instrument failed) · 8 unverified
Densely Connected Convolutional Networks 146 1 25 Aug 2016 community repositories only · 48 ran (of which 0 constructed an object rather than computing a result; 32 with no instrument failure: 2 honoured, 0 violated, 30 with no contract checked; 16 where Syntology's instrument failed) · 23 unverified (8 pointer-only for licence)
Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning 87 1 23 Feb 2016 1 ran (of which 0 constructed an object rather than computing a result; 0 with no instrument failure: 0 honoured, 0 violated, 0 with no contract checked; 1 where Syntology's instrument failed) · 0 unverified
Very Deep Convolutional Networks for Large-Scale Image Recognition 305 1 4 Sep 2014 81 ran (of which 0 constructed an object rather than computing a result; 69 with no instrument failure: 0 honoured, 0 violated, 69 with no contract checked; 12 where Syntology's instrument failed) · 41 unverified (8 pointer-only for licence)

Dataset loaders archive 2025-07-28

1 loader as listed in the archive; links are outbound and not re-checked here.

Tasks archive 2025-07-28

License archive 2025-07-28

CC BY-NC-SA

Modalities archive 2025-07-28

Languages archive 2025-07-28

Variants archive 2025-07-28

  • InDL

1 variant name, as the archive lists them.

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections