{"passport":{"unfragile":{"@version":"1.0","version":"2026-05","artifact":{"id":"imagenet-ilsvrc","slug":"imagenet-ilsvrc","name":"ImageNet (ILSVRC)","type":"dataset","url":"https://www.image-net.org/","page_url":"https://unfragile.ai/imagenet-ilsvrc","categories":["model-training","testing-quality"],"tags":[],"pricing":{"model":"free","free":true,"starting_price":null},"status":"active","verified":false},"capabilities":[{"id":"imagenet-ilsvrc__cap_0","uri":"capability://data.processing.analysis.large.scale.hierarchical.image.dataset.for.vision.model.pre.training","name":"large-scale hierarchical image dataset for vision model pre-training","description":"Provides 14.2 million images organized into 21,841 WordNet noun synsets with human-verified labels, enabling researchers to pre-train deep convolutional neural networks at scale. Images are sourced from the web and indexed by synset identifier, allowing models to learn visual representations across diverse object categories before fine-tuning on downstream tasks. The hierarchical WordNet structure maps synonym sets to image collections, creating a taxonomy-aware training corpus that supports both flat classification and hierarchical learning approaches.","intents":["Pre-train a vision model on large-scale labeled image data before fine-tuning on a domain-specific task","Access a standardized, publicly-available dataset for reproducible computer vision research","Train models that can classify objects across 1,000 or 21,841 semantic categories","Leverage transfer learning by using ImageNet-pretrained weights as initialization for custom models"],"best_for":["academic researchers in computer vision and deep learning","teams building production vision models who need strong initialization weights","educators teaching CNN architectures and transfer learning concepts","non-commercial organizations conducting image classification research"],"limitations":["Non-commercial use restriction prohibits direct commercial deployment or monetization of models trained on ImageNet","Image distribution is uneven across synsets (goal is ~1,000 per synset but variance exists), creating class imbalance","Web-sourced images have variable quality and availability; links may become stale over time","Limited to noun concepts only (80,000+ of 100,000+ WordNet synsets); excludes verbs, adjectives, and abstract concepts","No temporal metadata; all images are static snapshots without temporal context or video sequences","Known demographic bias in person-related categories (September 2019 filtering effort documented)","Privacy concerns identified in March 2021 update; some person images may not have explicit consent"],"requires":["Account registration on image-net.org with institutional affiliation verification for non-commercial access","Agreement to non-commercial research and educational use terms","Sufficient local storage: ~150GB for full dataset or ~50GB for ILSVRC 2012 subset","Python 3.6+ with PyTorch, TensorFlow, or similar framework for loading and processing images","Understanding of image classification tasks and WordNet hierarchy structure"],"input_types":["JPEG/PNG images (format specifications not documented; inferred from web image sources)","WordNet synset identifiers (e.g., 'n02084442' for 'dog')"],"output_types":["Labeled image datasets organized by synset directory","Annotation metadata (format unspecified in documentation)","Pre-trained model weights (when used with training frameworks)"],"categories":["data-processing-analysis","model-training-dataset"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"imagenet-ilsvrc__cap_1","uri":"capability://data.processing.analysis.ilsvrc.competition.benchmark.subset.with.standardized.evaluation.metrics","name":"ilsvrc competition benchmark subset with standardized evaluation metrics","description":"Provides a curated 1,000-class subset of ImageNet (1.28M training images) with standardized test set and evaluation protocol that defined the ImageNet Large Scale Visual Recognition Challenge. The benchmark uses top-5 accuracy as the primary metric, where a prediction is correct if the true label appears in the model's top-5 ranked predictions. This subset became the de facto standard for evaluating CNN architectures from AlexNet (2012, 83.6% top-5) through modern models (99%+ top-5), establishing a reproducible evaluation framework that enabled direct comparison of architectural innovations.","intents":["Benchmark a new CNN architecture against historical baselines using standardized ILSVRC metrics","Reproduce results from published papers that report ImageNet top-5 accuracy","Compare model performance across different frameworks using the same test set","Evaluate transfer learning effectiveness by measuring accuracy on ILSVRC classification task"],"best_for":["computer vision researchers publishing CNN architecture papers","teams evaluating model performance against published baselines","educators demonstrating the progression of deep learning (AlexNet → ResNet → Vision Transformers)","practitioners validating that pre-trained models achieve expected accuracy before deployment"],"limitations":["Benchmark is effectively saturated: top-5 accuracy has plateaued at 99%+, limiting ability to distinguish state-of-the-art models","Top-5 metric is lenient; top-1 accuracy is more challenging but less commonly reported historically","1,000 classes are biased toward object categories; lacks fine-grained distinctions (e.g., dog breeds) compared to specialized datasets","Test set was updated in October 2019; results from papers using older test sets may not be directly comparable","Evaluation is single-label classification; images with multiple objects are labeled with only one primary category","No localization or segmentation annotations in standard ILSVRC subset (localization task exists but separate)"],"requires":["Access to ILSVRC 2012 test set (available via image-net.org or Kaggle)","Implementation of top-5 accuracy metric in evaluation code","Model trained or fine-tuned on ILSVRC training set (1.28M images, 1,000 classes)","Computational resources to run inference on ~50K test images"],"input_types":["JPEG/PNG images from ILSVRC test set","Model predictions (logits or probabilities for 1,000 classes)"],"output_types":["Top-5 accuracy score (0.0-1.0)","Top-1 accuracy score (0.0-1.0)","Per-class accuracy metrics"],"categories":["data-processing-analysis","benchmark-evaluation"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"imagenet-ilsvrc__cap_2","uri":"capability://data.processing.analysis.wordnet.aligned.hierarchical.category.taxonomy.for.semantic.organization","name":"wordnet-aligned hierarchical category taxonomy for semantic organization","description":"Maps images to 21,841 WordNet noun synsets, where each synset represents a concept defined by a set of synonymous words (e.g., synset 'n02084442' contains 'dog', 'canis familiaris', 'Canis familiaris'). The hierarchy is inherited from WordNet's is-a relationships, enabling multi-level semantic organization where 'dog' is a hyponym of 'canine', which is a hyponym of 'mammal', etc. This structure allows models to learn hierarchical representations and enables zero-shot classification through semantic similarity in the WordNet graph, distinguishing ImageNet from datasets organized by arbitrary category names.","intents":["Organize image collections using linguistic semantic relationships rather than arbitrary category names","Enable zero-shot classification by leveraging WordNet hierarchy (e.g., predicting unseen dog breeds via parent 'dog' category)","Train models that understand hierarchical relationships between object categories","Map predictions to semantic concepts that align with human language and knowledge bases"],"best_for":["researchers working on zero-shot or few-shot learning using semantic hierarchies","teams building knowledge-grounded vision systems that must align with linguistic ontologies","educators explaining how semantic relationships can improve model generalization","systems requiring interpretable category names grounded in human language"],"limitations":["Limited to noun concepts only (80,000+ of 100,000+ WordNet synsets); excludes verbs, adjectives, and abstract concepts","WordNet hierarchy is manually curated and may not reflect visual similarity (e.g., 'bat' (animal) and 'bat' (tool) are separate synsets but visually distinct)","Synset-to-image mapping is not one-to-one; images are assigned to single synsets despite potentially depicting multiple objects","WordNet hierarchy is static and does not evolve with emerging object categories or modern language usage","No documentation on how to programmatically access or traverse the WordNet hierarchy within ImageNet"],"requires":["Understanding of WordNet structure and synset identifiers (e.g., 'n02084442')","Access to WordNet library (NLTK in Python or standalone WordNet database)","Mapping between ImageNet synset IDs and WordNet synset IDs","Knowledge of is-a relationships in WordNet hierarchy"],"input_types":["WordNet synset identifiers (e.g., 'n02084442')","Image labels (synset IDs)"],"output_types":["Hierarchical category paths (e.g., 'dog' → 'canine' → 'mammal' → 'animal')","Semantic similarity scores between synsets","Zero-shot predictions via hierarchy traversal"],"categories":["data-processing-analysis","memory-knowledge"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"imagenet-ilsvrc__cap_3","uri":"capability://data.processing.analysis.web.sourced.image.collection.with.url.based.access.and.copyright.attribution","name":"web-sourced image collection with url-based access and copyright attribution","description":"ImageNet does not host image files directly; instead, it maintains an indexed database of URLs pointing to images on the public web, with human-verified labels and copyright information. The dataset provides URLs, synset IDs, and metadata rather than image files, allowing users to download images on-demand while respecting original copyright holders. This URL-based approach reduces storage burden on ImageNet infrastructure and distributes copyright responsibility to users, but introduces challenges with link rot (URLs becoming invalid over time) and requires users to respect original copyright terms.","intents":["Access a large-scale image dataset without downloading 150GB+ of files upfront","Respect copyright by downloading images directly from original sources rather than redistributed copies","Build custom subsets by selectively downloading images matching specific criteria","Maintain dataset freshness by re-downloading images from original URLs"],"best_for":["researchers with limited storage who need selective image access","teams concerned with copyright compliance and original source attribution","projects requiring custom subsets of ImageNet (e.g., specific synsets or quality tiers)","systems that can tolerate occasional missing images due to link rot"],"limitations":["Link rot: original URLs become invalid over time, making some images inaccessible (no statistics provided on current availability)","Download latency: fetching images on-demand is slower than pre-downloaded datasets","No guarantee of image availability: original websites may remove images, change URLs, or block automated access","Copyright responsibility: users must respect original copyright holders' terms; ImageNet provides attribution but not licenses","Network dependency: requires internet connectivity to download images; offline use requires pre-downloading","No batch download API documented; users must implement custom download scripts"],"requires":["Internet connectivity to download images from original URLs","HTTP client library (curl, wget, Python requests, etc.) to fetch images","Respect for original copyright holders and their terms of use","Handling of failed downloads due to link rot or access restrictions","Storage for downloaded images (150GB for full dataset, 50GB for ILSVRC subset)"],"input_types":["ImageNet URL index (synset ID, image URL, copyright info)","HTTP requests to original image URLs"],"output_types":["JPEG/PNG image files","Copyright and attribution metadata","Download status (success/failure/link rot)"],"categories":["data-processing-analysis","search-retrieval"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"imagenet-ilsvrc__cap_4","uri":"capability://data.processing.analysis.human.verified.image.to.synset.annotation.with.quality.control","name":"human-verified image-to-synset annotation with quality control","description":"ImageNet employs human annotators to verify that images correctly represent their assigned WordNet synsets, implementing a quality control process to ensure label accuracy. The annotation process involves multiple annotators per image and consensus-based verification, reducing label noise compared to automated web scraping. This human verification is critical for benchmark reliability—mislabeled images would corrupt model evaluation and make architectural comparisons unreliable. The quality control process is not fully documented, but the artifact mentions 'human-annotated and quality-controlled' images.","intents":["Access a dataset with verified labels suitable for rigorous model evaluation and benchmarking","Train models on high-quality annotations that reduce label noise and improve convergence","Conduct research that depends on label accuracy (e.g., studying model robustness to label noise)","Establish a reliable benchmark where label errors are minimized"],"best_for":["researchers requiring high-quality labels for rigorous benchmarking","teams training models where label noise significantly impacts performance","studies investigating model robustness or label quality effects","practitioners validating that benchmark results are not artifacts of label errors"],"limitations":["Quality control process is not fully documented; no published statistics on inter-annotator agreement or label error rates","Human annotation is subjective; some images may legitimately belong to multiple synsets but are assigned only one","Annotation quality may vary across synsets (e.g., fine-grained categories like dog breeds may have higher error rates)","No public error analysis or known label errors list; researchers cannot easily identify problematic images","Quality control was performed at dataset creation time (2009-2014); no ongoing verification for new images"],"requires":["Trust in ImageNet's annotation process (no independent verification available)","Understanding that some label errors likely exist despite quality control","Awareness of potential systematic biases in human annotation (e.g., annotator demographics)"],"input_types":["Raw web images","WordNet synset definitions"],"output_types":["Verified image-synset pairs","Annotation metadata (annotator IDs, confidence scores if available)"],"categories":["data-processing-analysis","safety-moderation"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"imagenet-ilsvrc__cap_5","uri":"capability://safety.moderation.non.commercial.research.license.with.institutional.access.control","name":"non-commercial research license with institutional access control","description":"ImageNet restricts access to non-commercial research and educational use through a login-based access control system that requires institutional affiliation verification. Users must agree to terms prohibiting commercial deployment, monetization, or use of models trained on ImageNet. This licensing model protects ImageNet's legal position regarding copyright of original images (which ImageNet does not own) while enabling academic research. Access is granted 'under certain conditions and terms' that are not fully detailed in public documentation, creating ambiguity about what constitutes permitted use.","intents":["Access ImageNet as an academic researcher without commercial restrictions","Verify institutional affiliation to qualify for non-commercial research access","Understand legal constraints on model deployment and commercialization","Ensure compliance with ImageNet's terms before publishing research"],"best_for":["academic researchers at universities and research institutions","non-profit organizations conducting educational research","educators teaching computer vision and deep learning","teams prototyping models for research purposes before commercialization"],"limitations":["Non-commercial restriction prohibits direct commercial deployment of models trained on ImageNet","Institutional affiliation requirement excludes independent researchers and small companies","Terms are vague: 'non-commercial research and educational purposes' is not precisely defined (e.g., is fine-tuning for a commercial product permitted?)","No clear guidance on whether models pre-trained on ImageNet can be used in commercial products if fine-tuned on other data","Access control is manual; approval process timeline is not documented","No public appeals process if access is denied"],"requires":["Institutional email address or affiliation verification","Account creation on image-net.org","Agreement to non-commercial use terms","Understanding of legal implications of non-commercial restriction"],"input_types":["Institutional affiliation information","User agreement acceptance"],"output_types":["Access credentials (login)","Download links to dataset"],"categories":["safety-moderation","tool-use-integration"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"imagenet-ilsvrc__cap_6","uri":"capability://memory.knowledge.transfer.learning.initialization.via.pre.trained.model.weights","name":"transfer learning initialization via pre-trained model weights","description":"ImageNet enables transfer learning by serving as the standard pre-training dataset for vision models. Researchers train CNNs on ImageNet's 1.28M images (ILSVRC) or full 14.2M images, then release pre-trained weights that practitioners use as initialization for downstream tasks. This approach leverages ImageNet's scale and diversity to learn general-purpose visual features (edges, textures, object parts) that transfer to specialized domains. Modern frameworks (PyTorch, TensorFlow) provide ImageNet-pretrained weights for standard architectures (ResNet, VGG, Vision Transformers), making transfer learning a standard practice.","intents":["Initialize a model with ImageNet-pretrained weights to accelerate training on a downstream task","Reduce training time and data requirements by leveraging features learned from 1.28M images","Improve performance on small datasets by using ImageNet pre-training as regularization","Establish a common baseline for comparing downstream task performance across papers"],"best_for":["practitioners with limited labeled data for a specific vision task (medical imaging, satellite imagery, etc.)","teams with constrained compute budgets who need faster training","researchers comparing models on downstream tasks using standardized initialization","educators demonstrating transfer learning effectiveness"],"limitations":["Domain shift: ImageNet features may not transfer well to domains with different visual characteristics (e.g., medical imaging, microscopy)","Fine-tuning hyperparameters are critical; poor learning rates or regularization can degrade pre-trained weights","Architectural mismatch: pre-trained weights are specific to architecture (ResNet-50 weights don't transfer to Vision Transformer)","Computational cost of pre-training is amortized across many downstream tasks; individual practitioners don't see the 1.28M-image training cost","Non-commercial restriction on ImageNet may create legal ambiguity for commercial products using ImageNet-pretrained weights"],"requires":["Pre-trained model weights (available from PyTorch, TensorFlow, or other framework hubs)","Downstream task dataset (can be small; transfer learning is most effective with <10K images)","Fine-tuning code that adjusts learning rates and regularization for the downstream task","Understanding of transfer learning principles and when it is effective"],"input_types":["Pre-trained model weights (PyTorch .pth, TensorFlow .h5, etc.)","Downstream task images and labels"],"output_types":["Fine-tuned model weights","Downstream task predictions","Performance metrics (accuracy, F1, etc.)"],"categories":["memory-knowledge","code-generation-editing"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"imagenet-ilsvrc__cap_7","uri":"capability://data.processing.analysis.multi.label.and.fine.grained.category.support.for.specialized.vision.tasks","name":"multi-label and fine-grained category support for specialized vision tasks","description":"While standard ILSVRC uses single-label classification, ImageNet's full 21,841-synset structure includes fine-grained categories (e.g., dog breeds: 'Chihuahua', 'German Shepherd', 'Poodle') that enable specialized vision tasks beyond basic object recognition. The hierarchical structure allows models to learn both coarse-grained (dog) and fine-grained (Chihuahua) distinctions, supporting applications like species identification, product recognition, and medical imaging. However, the single-label-per-image constraint limits multi-label learning (e.g., images with multiple objects), and fine-grained categories have fewer images per synset, creating class imbalance.","intents":["Train fine-grained classification models (e.g., dog breed recognition, bird species identification)","Build hierarchical classifiers that predict both coarse and fine-grained categories","Leverage ImageNet's fine-grained synsets for specialized domains (e.g., product recognition, medical imaging)","Study how models learn hierarchical visual distinctions"],"best_for":["teams building specialized vision systems (e.g., wildlife monitoring, product catalogs, medical diagnosis)","researchers studying fine-grained visual recognition and hierarchical classification","practitioners needing pre-trained weights for fine-grained categories","educators demonstrating hierarchical learning and multi-level classification"],"limitations":["Single-label constraint: images are assigned to one synset despite potentially depicting multiple objects; limits multi-label learning","Class imbalance: fine-grained categories have fewer images per synset (goal is ~1,000 but variance is high), making training difficult","Limited fine-grained diversity: ImageNet's fine-grained categories focus on animals and objects; lacks fine-grained distinctions for other domains","Hierarchical structure is inherited from WordNet, which may not reflect visual similarity (e.g., 'bat' (animal) and 'bat' (tool) are separate)","No explicit multi-label annotations; images with multiple objects are labeled with only one primary category"],"requires":["Understanding of hierarchical classification and fine-grained visual recognition","Handling of class imbalance (e.g., weighted loss, oversampling, data augmentation)","Fine-tuning strategy that preserves pre-trained features while adapting to fine-grained distinctions","Evaluation metrics appropriate for hierarchical classification (e.g., hierarchical precision/recall)"],"input_types":["Images from fine-grained ImageNet synsets","Hierarchical category labels (coarse and fine-grained)"],"output_types":["Fine-grained predictions (e.g., 'Chihuahua' instead of 'dog')","Hierarchical classification scores","Confidence scores for each level of hierarchy"],"categories":["data-processing-analysis","memory-knowledge"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"imagenet-ilsvrc__cap_8","uri":"capability://safety.moderation.privacy.aware.person.category.filtering.and.demographic.balancing","name":"privacy-aware person category filtering and demographic balancing","description":"ImageNet's person-related synsets (e.g., 'person', 'child', 'athlete') contain images of real people, raising privacy and demographic bias concerns. In September 2019, ImageNet published a research update on 'filtering and balancing the ImageNet person subtree,' and in March 2021, a paper on 'privacy preservation' was released, indicating efforts to address privacy issues. The specific filtering and balancing approach is not detailed in available documentation, but likely involves removing images without explicit consent and rebalancing demographic representation across person categories. This capability reflects growing awareness of privacy and fairness issues in large-scale datasets.","intents":["Use ImageNet for person-related tasks while respecting privacy of individuals in images","Train models on demographically balanced person categories to reduce bias","Understand privacy implications of large-scale image datasets containing real people","Comply with privacy regulations (e.g., GDPR) when using ImageNet for research"],"best_for":["researchers studying fairness and bias in vision models","teams building person-related vision systems (face recognition, pose estimation) with privacy concerns","organizations subject to privacy regulations (GDPR, CCPA) using ImageNet","educators discussing privacy and ethics in machine learning"],"limitations":["Privacy filtering approach is not fully documented; unclear which images were removed or why","Demographic balancing methodology is not published; unclear how 'balance' is defined or measured","No public list of removed images or filtering criteria; researchers cannot verify privacy compliance","Demographic categories used for balancing are not specified (e.g., age, gender, ethnicity, skin tone)","Privacy concerns remain: even filtered person images may not have explicit consent from individuals","Demographic bias may persist despite balancing efforts; no published fairness metrics"],"requires":["Awareness of privacy and fairness issues in person-related datasets","Understanding of demographic bias and its impact on model performance","Compliance with privacy regulations if using person images for commercial purposes","Ethical review if conducting research on sensitive person categories"],"input_types":["Person-related ImageNet synsets (filtered and balanced)","Demographic labels (if available)"],"output_types":["Filtered person images","Demographic distribution statistics","Privacy and fairness metrics"],"categories":["safety-moderation","data-processing-analysis"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"imagenet-ilsvrc__headline","uri":"capability://image.visual.large.scale.image.dataset.for.training.and.benchmarking.deep.learning.models","name":"large-scale image dataset for training and benchmarking deep learning models","description":"ImageNet is a foundational dataset containing over 14 million images organized into 21,841 categories, widely used for training and benchmarking deep learning models in computer vision tasks.","intents":["best image dataset for deep learning","image dataset for training CNNs","benchmark dataset for object recognition","large-scale dataset for image classification","free image dataset for research"],"best_for":["deep learning researchers","computer vision practitioners"],"limitations":["copyright restrictions","static dataset"],"requires":["basic understanding of machine learning"],"input_types":["images"],"output_types":["trained models","evaluation metrics"],"categories":["image-visual"],"confidence":0.5,"matches":0,"success_rate":0}],"trust":{"score":57,"verified":false,"data_access_risk":"high","permissions":["Account registration on image-net.org with institutional affiliation verification for non-commercial access","Agreement to non-commercial research and educational use terms","Sufficient local storage: ~150GB for full dataset or ~50GB for ILSVRC 2012 subset","Python 3.6+ with PyTorch, TensorFlow, or similar framework for loading and processing images","Understanding of image classification tasks and WordNet hierarchy structure","Access to ILSVRC 2012 test set (available via image-net.org or Kaggle)","Implementation of top-5 accuracy metric in evaluation code","Model trained or fine-tuned on ILSVRC training set (1.28M images, 1,000 classes)","Computational resources to run inference on ~50K test images","Understanding of WordNet structure and synset identifiers (e.g., 'n02084442')"],"failure_modes":["Non-commercial use restriction prohibits direct commercial deployment or monetization of models trained on ImageNet","Image distribution is uneven across synsets (goal is ~1,000 per synset but variance exists), creating class imbalance","Web-sourced images have variable quality and availability; links may become stale over time","Limited to noun concepts only (80,000+ of 100,000+ WordNet synsets); excludes verbs, adjectives, and abstract concepts","No temporal metadata; all images are static snapshots without temporal context or video sequences","Known demographic bias in person-related categories (September 2019 filtering effort documented)","Privacy concerns identified in March 2021 update; some person images may not have explicit consent","Benchmark is effectively saturated: top-5 accuracy has plateaued at 99%+, limiting ability to distinguish state-of-the-art models","Top-5 metric is lenient; top-1 accuracy is more challenging but less commonly reported historically","1,000 classes are biased toward object categories; lacks fine-grained distinctions (e.g., dog breeds) compared to specialized datasets","builder identity is not verified yet","no observed match outcomes yet"],"rank_breakdown":{"adoption":0.7,"quality":0.8500000000000001,"ecosystem":0.39999999999999997,"match_graph":0.25,"freshness":0.75,"weights":{"adoption":0.3,"quality":0.25,"ecosystem":0.1,"match_graph":0.3,"freshness":0.05}},"observed_outcomes":{"matches":0,"success_rate":0,"avg_confidence":0,"top_intents":[],"last_matched_at":null},"maintenance":{"status":"active","updated_at":"2026-05-24T12:16:23.327Z","last_scraped_at":null,"last_commit":null},"community":{"stars":null,"forks":null,"weekly_downloads":null,"model_downloads":null,"model_likes":null}},"distribution":{"claim_url":"https://unfragile.ai/submit?claim=imagenet-ilsvrc","compare_url":"https://unfragile.ai/compare?artifact=imagenet-ilsvrc"}},"signature":"FvGRiyRT3NXiD10nNE0zEEV9v1BB7SuC72M3a7NogHC4f/YF0DKKF5zNasU4MSVzD8Ov2vzgL0KRS9ZrPkMsBw==","signedAt":"2026-06-22T01:58:40.721Z","signedBy":"unfragile.ai","version":1},"_links":{"self":"https://unfragile.ai/api/v1/passport/imagenet-ilsvrc","artifact":"https://unfragile.ai/imagenet-ilsvrc","verify":"https://unfragile.ai/api/v1/verify?slug=imagenet-ilsvrc","publicKey":"https://unfragile.ai/api/v1/trust-passport-public-key","spec":"https://unfragile.ai/trust","schema":"https://unfragile.ai/schema.json","docs":"https://unfragile.ai/docs"}}