{"passport":{"unfragile":{"@version":"1.0","version":"2026-05","artifact":{"id":"hf-model-shi-labs--oneformer_ade20k_swin_large","slug":"shi-labs--oneformer_ade20k_swin_large","name":"oneformer_ade20k_swin_large","type":"model","url":"https://huggingface.co/shi-labs/oneformer_ade20k_swin_large","page_url":"https://unfragile.ai/shi-labs--oneformer_ade20k_swin_large","categories":["image-generation"],"tags":["transformers","pytorch","oneformer","vision","image-segmentation","universal-image-segmentation","dataset:scene_parse_150","arxiv:2211.06220","license:mit","endpoints_compatible","deploy:azure","region:us"],"pricing":{"model":"open_source","free":true,"starting_price":null},"status":"active","verified":false},"capabilities":[{"id":"hf-model-shi-labs--oneformer_ade20k_swin_large__cap_0","uri":"capability://image.visual.unified.panoptic.semantic.instance.segmentation","name":"unified-panoptic-semantic-instance-segmentation","description":"Performs simultaneous panoptic, semantic, and instance segmentation on images using a unified transformer-based architecture. Leverages Swin Transformer backbone with deformable cross-attention mechanisms to process multi-scale visual features and generate dense pixel-level predictions across all three segmentation tasks in a single forward pass, eliminating the need for task-specific model variants.","intents":["I need to segment both stuff (sky, wall) and things (person, car) in a single model inference","I want to get semantic class labels, instance boundaries, and panoptic masks without running multiple models","I need to understand scene composition with unified predictions across different segmentation paradigms"],"best_for":["computer vision researchers building multi-task segmentation pipelines","autonomous systems engineers requiring comprehensive scene understanding","teams deploying edge models where model count and latency are constrained"],"limitations":["Trained exclusively on ADE20K dataset (150 semantic classes) — zero-shot transfer to other domains requires fine-tuning","Inference latency ~500-800ms on GPU for 512x512 images; CPU inference impractical for real-time applications","Memory footprint ~1.3GB for model weights; requires GPU with minimum 4GB VRAM for batch processing","Performance degrades on images with extreme aspect ratios or very small objects (<32 pixels)"],"requires":["PyTorch 1.9+","transformers library 4.25+","CUDA 11.0+ for GPU acceleration (optional but strongly recommended)","Pillow for image preprocessing","numpy for output tensor manipulation"],"input_types":["RGB images (PIL Image, numpy array, or file path)","Images with arbitrary resolution (internally resized to 512x512 or 1024x1024)"],"output_types":["panoptic segmentation map (H×W integer tensor with unique IDs per instance)","semantic segmentation map (H×W integer tensor with class indices 0-149)","instance segmentation map (H×W integer tensor with instance IDs)","class probability logits (H×W×150 float tensor)"],"categories":["image-visual","scene-understanding"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hf-model-shi-labs--oneformer_ade20k_swin_large__cap_1","uri":"capability://image.visual.swin.transformer.hierarchical.feature.extraction","name":"swin-transformer-hierarchical-feature-extraction","description":"Extracts multi-scale hierarchical visual features using Swin Transformer backbone with shifted window attention mechanism. Processes images through 4 stages with progressive spatial downsampling (4×, 8×, 16×, 32×) while maintaining computational efficiency through local window-based self-attention instead of global quadratic attention, producing feature pyramids compatible with dense prediction heads.","intents":["I need efficient multi-scale feature extraction without quadratic memory complexity","I want to leverage pre-trained vision transformer weights for faster convergence","I need to extract features at multiple resolutions for dense prediction tasks"],"best_for":["teams building dense prediction models (segmentation, depth estimation) with GPU memory constraints","researchers requiring interpretable attention patterns via shifted window visualization","production systems where inference latency must be <1 second on consumer GPUs"],"limitations":["Shifted window attention requires careful padding/masking — incompatible with some quantization schemes","Feature resolution limited to input image size; very high-resolution inputs (>2048×2048) cause memory overflow","Swin-Large has 196M parameters — requires careful optimization for mobile/edge deployment","Attention patterns are local (window-based) — may miss long-range dependencies critical for some scenes"],"requires":["PyTorch 1.9+","timm library 0.6.0+ for Swin backbone implementation","CUDA 11.0+ recommended for efficient window attention kernels"],"input_types":["RGB images (arbitrary resolution, internally padded to multiple of 32)"],"output_types":["4-level feature pyramid: C2 (H/4×W/4×96), C3 (H/8×W/8×192), C4 (H/16×W/16×384), C5 (H/32×W/32×768)"],"categories":["image-visual","feature-extraction"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hf-model-shi-labs--oneformer_ade20k_swin_large__cap_10","uri":"capability://image.visual.ade20k.dataset.finetuning.compatibility","name":"ade20k-dataset-finetuning-compatibility","description":"Provides pretrained weights optimized for ADE20K dataset (150 semantic classes, 20K training images) with training recipes and hyperparameters documented. Enables efficient fine-tuning on custom datasets by leveraging learned feature representations and class embeddings.","intents":["I want to fine-tune the model on a custom dataset without training from scratch","I need to adapt the model to a different set of semantic classes","I want to understand the training setup used for ADE20K pretraining"],"best_for":["teams with limited labeled data (1K-5K images) for custom segmentation tasks","researchers studying transfer learning from ADE20K to other domains","applications where ADE20K classes partially overlap with target domain"],"limitations":["Fine-tuning requires modifying the 150-class output head — cannot reuse pretrained class embeddings for new classes","Transfer learning effectiveness depends on domain similarity — fine-tuning on outdoor scenes (COCO) may require 50%+ of original training data","Catastrophic forgetting risk — fine-tuning on small datasets may degrade performance on ADE20K classes","No documented fine-tuning recipes — requires manual hyperparameter tuning (learning rate, batch size, augmentation)"],"requires":["PyTorch 1.9+","transformers 4.25+","Custom dataset with pixel-level annotations","Training framework (e.g., PyTorch Lightning, mmdetection)"],"input_types":["Custom dataset with RGB images and semantic segmentation masks"],"output_types":["Fine-tuned model checkpoint","Validation metrics (mIoU, mAP, PQ)"],"categories":["image-visual","transfer-learning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hf-model-shi-labs--oneformer_ade20k_swin_large__cap_11","uri":"capability://tool.use.integration.mit.license.open.source.deployment","name":"mit-license-open-source-deployment","description":"Released under MIT license enabling unrestricted commercial and research use, modification, and redistribution. Model weights and code are publicly available on Hugging Face Model Hub with no licensing restrictions or attribution requirements beyond standard MIT terms.","intents":["I want to use this model in a commercial product without licensing fees","I need to modify and redistribute the model for my use case","I want to ensure there are no legal restrictions on deployment"],"best_for":["commercial teams building products with segmentation capabilities","open-source projects requiring permissive licensing","researchers publishing models and code without restrictions"],"limitations":["MIT license provides no warranty — model performance issues are not covered by support","No patent protection — commercial use may expose users to patent claims from other parties","Attribution not required but recommended for academic integrity","Model weights are public — no confidentiality or competitive advantage from using this model"],"requires":["Acceptance of MIT license terms","No additional licensing or registration"],"input_types":["MIT license agreement"],"output_types":["Unrestricted usage rights"],"categories":["tool-use-integration","licensing"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hf-model-shi-labs--oneformer_ade20k_swin_large__cap_12","uri":"capability://tool.use.integration.huggingface.endpoints.cloud.deployment","name":"huggingface-endpoints-cloud-deployment","description":"Compatible with Hugging Face Inference Endpoints for serverless cloud deployment. Model can be deployed as a managed endpoint with automatic scaling, monitoring, and API access without managing infrastructure.","intents":["I want to deploy the model to the cloud without managing servers","I need automatic scaling based on inference load","I want to access the model via REST API from any application"],"best_for":["teams without DevOps expertise wanting quick cloud deployment","applications with variable inference load requiring auto-scaling","web applications needing REST API access to segmentation"],"limitations":["Hugging Face Endpoints pricing is per-hour of compute — continuous deployment is expensive vs self-hosted","Inference latency includes network overhead — 100-200ms added to model latency for API calls","Cold start latency (first request after scaling down) can be 5-10 seconds","Limited customization — cannot modify inference code or add custom preprocessing without using custom Docker images"],"requires":["Hugging Face account with billing enabled","Endpoint creation via Hugging Face UI or API","API key for authentication"],"input_types":["Image file (base64 encoded or multipart upload)","Image URL"],"output_types":["JSON response with segmentation maps (base64 encoded) and metadata"],"categories":["tool-use-integration","cloud-deployment"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hf-model-shi-labs--oneformer_ade20k_swin_large__cap_2","uri":"capability://image.visual.deformable.cross.attention.fusion","name":"deformable-cross-attention-fusion","description":"Fuses multi-scale features using deformable cross-attention modules that learn to attend to task-relevant spatial regions dynamically. Each attention head learns offset predictions to sample features from adaptive 2D positions rather than fixed grids, enabling the model to focus on semantically important regions (object boundaries, fine details) while ignoring background noise.","intents":["I need the model to focus on object boundaries and fine details rather than uniform feature sampling","I want to handle objects at varying scales without explicit multi-scale processing","I need to reduce computational cost of attention by sampling sparse spatial locations"],"best_for":["applications with objects spanning multiple scales (aerial imagery, medical imaging with variable anatomy sizes)","systems requiring interpretable attention — deformable offsets can be visualized to understand model focus","resource-constrained deployments where sparse attention reduces memory and compute"],"limitations":["Deformable offset learning is non-differentiable in some frameworks — requires custom CUDA kernels for efficiency","Learned offsets can become unstable during early training — requires careful initialization and learning rate scheduling","Visualization of deformable attention is complex — harder to debug than standard attention patterns","Performance gains diminish on uniform, well-structured scenes (synthetic data, clean datasets)"],"requires":["PyTorch 1.9+ with CUDA support for deformable convolution kernels","torchvision 0.10+ for deformable attention implementations","Custom CUDA compilation may be needed for optimal performance"],"input_types":["Multi-scale feature maps from backbone (C2-C5 pyramid)"],"output_types":["Fused feature maps with same spatial dimensions as input","Attention offset maps (H×W×2×num_heads) showing learned sampling positions"],"categories":["image-visual","attention-mechanism"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hf-model-shi-labs--oneformer_ade20k_swin_large__cap_3","uri":"capability://image.visual.task.conditioned.query.generation","name":"task-conditioned-query-generation","description":"Generates task-specific query embeddings (panoptic, semantic, instance) that condition a shared transformer decoder to produce task-appropriate outputs. Each task has learnable query tokens that are concatenated with image features and processed through cross-attention layers, allowing the same decoder weights to produce different segmentation outputs based on task conditioning.","intents":["I want to run multiple segmentation tasks without maintaining separate model checkpoints","I need to switch between panoptic, semantic, and instance segmentation at inference time","I want to understand how task information flows through the model architecture"],"best_for":["production systems requiring flexible task switching without model reloading","research teams studying task relationships and shared representations","applications where task requirements change dynamically (e.g., user-selected segmentation mode)"],"limitations":["Task queries must be learned jointly — fine-tuning for new tasks requires retraining the entire model","Query embeddings are fixed-size (256-dim) — may bottleneck information flow for complex task specifications","No explicit mechanism to prevent task interference — panoptic queries may inadvertently influence semantic outputs","Task switching at inference has negligible cost, but training requires careful balancing of task losses to prevent one task from dominating"],"requires":["PyTorch 1.9+","transformers 4.25+ for decoder implementation"],"input_types":["Task identifier (string: 'panoptic', 'semantic', or 'instance')","Image features from backbone (C2-C5 pyramid)"],"output_types":["Task-specific segmentation maps (H×W integer or float tensors)","Query attention weights (num_queries×H×W showing which image regions each query attends to)"],"categories":["image-visual","conditional-generation"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hf-model-shi-labs--oneformer_ade20k_swin_large__cap_4","uri":"capability://image.visual.ade20k.150.class.semantic.prediction","name":"ade20k-150-class-semantic-prediction","description":"Predicts semantic class labels from a fixed vocabulary of 150 ADE20K scene categories (wall, floor, ceiling, person, car, tree, etc.) using learned class embeddings and cross-entropy loss. The model outputs per-pixel logits over 150 classes, which are converted to class predictions via argmax or softmax for confidence scores.","intents":["I need to identify what semantic category each pixel belongs to (stuff and things)","I want to understand scene composition in terms of standard ADE20K categories","I need confidence scores for each pixel's class prediction"],"best_for":["scene understanding applications built on ADE20K (indoor scene analysis, robotics navigation)","research projects studying semantic segmentation on diverse indoor/outdoor scenes","applications where the 150 ADE20K classes cover the target domain"],"limitations":["Fixed to 150 classes — cannot predict novel classes outside ADE20K vocabulary without retraining","Class imbalance in ADE20K (some classes like 'wall' are 10× more frequent than 'lamp') causes biased predictions toward common classes","Performance varies significantly across classes: 85+ mIoU on common classes (wall, floor, person) but <40 mIoU on rare classes (<0.1% of pixels)","Requires fine-tuning for domain shift (e.g., outdoor scenes, synthetic data) — zero-shot transfer accuracy drops 15-25 mIoU"],"requires":["PyTorch 1.9+","transformers 4.25+","Knowledge of ADE20K class mapping (150 class indices to names)"],"input_types":["RGB images (arbitrary resolution, internally resized)"],"output_types":["Semantic segmentation map (H×W integer tensor with class indices 0-149)","Class logits (H×W×150 float tensor)","Class confidence scores (H×W×150 softmax probabilities)"],"categories":["image-visual","semantic-segmentation"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hf-model-shi-labs--oneformer_ade20k_swin_large__cap_5","uri":"capability://image.visual.instance.boundary.aware.segmentation","name":"instance-boundary-aware-segmentation","description":"Segments individual object instances by predicting instance masks that respect object boundaries and spatial separation. Uses instance queries (100-200 learnable embeddings) that compete during decoding to assign pixels to distinct instances, with boundary refinement through mask refinement modules that sharpen instance edges.","intents":["I need to identify and separate individual objects (e.g., each person, each car) in a scene","I want precise instance boundaries without post-processing (e.g., watershed, connected components)","I need instance-level features for downstream tasks (tracking, counting, attribute prediction)"],"best_for":["object detection and tracking pipelines requiring instance masks","robotics applications needing to grasp or manipulate individual objects","video analysis systems tracking objects across frames using instance consistency"],"limitations":["Fixed number of instance queries (100-200) — cannot handle scenes with >200 objects without modification","Instance assignment is ambiguous for overlapping or touching objects — may merge nearby instances or split single objects","Boundary precision degrades on small objects (<32 pixels) due to feature map resolution","No explicit instance ID consistency across frames — requires post-processing for video instance tracking"],"requires":["PyTorch 1.9+","transformers 4.25+","Post-processing utilities for instance mask refinement (optional)"],"input_types":["RGB images (arbitrary resolution)"],"output_types":["Instance segmentation map (H×W integer tensor with unique instance IDs)","Instance masks (num_instances×H×W binary masks)","Instance confidence scores (num_instances float tensor)"],"categories":["image-visual","instance-segmentation"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hf-model-shi-labs--oneformer_ade20k_swin_large__cap_6","uri":"capability://image.visual.panoptic.segmentation.stuff.things.unification","name":"panoptic-segmentation-stuff-things-unification","description":"Produces panoptic segmentation by unifying semantic (stuff) and instance (things) predictions into a single output where each pixel has a unique ID encoding both class and instance. Implements a merging algorithm that assigns instance IDs to stuff classes and instance-level IDs to thing classes, resolving overlaps through confidence-based prioritization.","intents":["I need a single unified segmentation output combining stuff (background) and things (objects)","I want to measure scene composition with a single metric (PQ) rather than separate mIoU and AP","I need to avoid post-processing overlaps between semantic and instance predictions"],"best_for":["scene understanding systems requiring holistic scene representation","autonomous driving perception stacks (stuff = road/sky, things = vehicles/pedestrians)","robotics applications needing complete scene segmentation for navigation and manipulation"],"limitations":["Panoptic quality (PQ) metric is complex — requires exact instance ID matching, making small errors heavily penalized","Stuff-things merging is heuristic-based — overlapping predictions are resolved by confidence thresholding, which can be suboptimal","Performance is bounded by the weaker of semantic or instance predictions — if semantic is 50 mIoU and instance is 30 AP, panoptic PQ will be low","No explicit mechanism to enforce stuff-things consistency — semantic and instance predictions may contradict (e.g., predicting 'person' as stuff)"],"requires":["PyTorch 1.9+","transformers 4.25+","Panoptic quality evaluation utilities (pycocotools or custom implementation)"],"input_types":["RGB images (arbitrary resolution)"],"output_types":["Panoptic segmentation map (H×W integer tensor with unique IDs: (class_id << 16) | instance_id)","Panoptic quality score (float 0-100)","Per-class panoptic quality breakdown"],"categories":["image-visual","panoptic-segmentation"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hf-model-shi-labs--oneformer_ade20k_swin_large__cap_7","uri":"capability://image.visual.batch.inference.with.variable.resolution","name":"batch-inference-with-variable-resolution","description":"Processes multiple images of different resolutions in a single batch by padding to a common size and tracking original dimensions for output resizing. Implements efficient batching logic that groups images by resolution to minimize padding overhead, with automatic output resizing to original image dimensions.","intents":["I want to process multiple images efficiently without resizing them to a fixed resolution","I need to handle images with different aspect ratios in a single batch","I want to minimize memory overhead from padding while maintaining batch efficiency"],"best_for":["production inference pipelines processing diverse image sources (web uploads, sensor streams)","batch processing systems (e.g., video frame processing) where resolution varies","applications requiring high throughput where batching is critical for GPU utilization"],"limitations":["Padding to common resolution increases memory usage — a batch of 1024×768 and 512×512 images requires padding to 1024×768, wasting ~25% memory","Output resizing introduces interpolation artifacts — bilinear resizing can blur fine instance boundaries","Batch size is limited by largest image in batch — a single 2048×2048 image forces all others to pad to that size","Variable resolution batching adds complexity — requires careful tracking of original dimensions and resizing logic"],"requires":["PyTorch 1.9+","torchvision for efficient image resizing","Batch processing framework (e.g., DataLoader with custom collate function)"],"input_types":["List of RGB images with arbitrary resolutions"],"output_types":["Batched segmentation maps resized to original image dimensions","Batched logits/confidence scores"],"categories":["image-visual","batch-processing"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hf-model-shi-labs--oneformer_ade20k_swin_large__cap_8","uri":"capability://tool.use.integration.huggingface.transformers.integration","name":"huggingface-transformers-integration","description":"Integrates with Hugging Face transformers library via AutoModel and AutoImageProcessor APIs, enabling one-line model loading and inference. Provides standardized preprocessing (image normalization, resizing) and postprocessing (output tensor conversion) through the transformers ecosystem.","intents":["I want to load the model with a single line of code without custom initialization","I need standard preprocessing that matches the model's training setup","I want to use the model in transformers-based pipelines (e.g., transformers.pipeline('image-segmentation'))"],"best_for":["developers using transformers library for other NLP/vision tasks","teams building modular ML pipelines with standardized interfaces","researchers prototyping models quickly without custom loading code"],"limitations":["Transformers integration requires specific model config format — custom modifications require forking the model card","ImageProcessor standardization may not match optimal preprocessing for specific use cases","Pipeline API abstracts away low-level control — difficult to customize inference (e.g., batch size, device placement)","Version compatibility issues — older transformers versions may not support this model"],"requires":["transformers 4.25+","PyTorch 1.9+","Pillow for image loading"],"input_types":["Image file path (string)","PIL Image object","numpy array (H×W×3)"],"output_types":["transformers.image_processing_utils.ImageFeatureExtractionMixin output","Segmentation maps in transformers standard format"],"categories":["tool-use-integration","model-loading"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"hf-model-shi-labs--oneformer_ade20k_swin_large__cap_9","uri":"capability://tool.use.integration.pytorch.checkpoint.loading.and.inference","name":"pytorch-checkpoint-loading-and-inference","description":"Loads pretrained weights from PyTorch checkpoint files (.pt, .pth) and performs inference on GPU or CPU. Implements state_dict compatibility checking and automatic device placement, with support for mixed-precision inference (fp16) for reduced memory usage.","intents":["I want to load pretrained weights and run inference without training","I need to run the model on GPU for speed or CPU for compatibility","I want to use mixed-precision inference to reduce memory usage"],"best_for":["production inference systems with fixed model weights","edge devices with limited memory (using fp16 quantization)","teams deploying models without fine-tuning"],"limitations":["Checkpoint loading requires exact architecture match — cannot load weights into modified model architectures","Mixed-precision (fp16) inference may reduce accuracy by 0.5-1 mIoU due to numerical precision loss","CPU inference is 10-20× slower than GPU — impractical for real-time applications","No built-in checkpoint versioning — loading old checkpoints may fail if model architecture changed"],"requires":["PyTorch 1.9+","CUDA 11.0+ for GPU inference (optional)","Checkpoint file (.pt or .pth) with matching architecture"],"input_types":["Checkpoint file path (string)","Model architecture instance (torch.nn.Module)"],"output_types":["Loaded model with pretrained weights","Inference outputs (segmentation maps, logits)"],"categories":["tool-use-integration","model-loading"],"confidence":0.5,"matches":0,"success_rate":0}],"trust":{"score":44,"verified":false,"data_access_risk":"high","permissions":["PyTorch 1.9+","transformers library 4.25+","CUDA 11.0+ for GPU acceleration (optional but strongly recommended)","Pillow for image preprocessing","numpy for output tensor manipulation","timm library 0.6.0+ for Swin backbone implementation","CUDA 11.0+ recommended for efficient window attention kernels","transformers 4.25+","Custom dataset with pixel-level annotations","Training framework (e.g., PyTorch Lightning, mmdetection)"],"failure_modes":["Trained exclusively on ADE20K dataset (150 semantic classes) — zero-shot transfer to other domains requires fine-tuning","Inference latency ~500-800ms on GPU for 512x512 images; CPU inference impractical for real-time applications","Memory footprint ~1.3GB for model weights; requires GPU with minimum 4GB VRAM for batch processing","Performance degrades on images with extreme aspect ratios or very small objects (<32 pixels)","Shifted window attention requires careful padding/masking — incompatible with some quantization schemes","Feature resolution limited to input image size; very high-resolution inputs (>2048×2048) cause memory overflow","Swin-Large has 196M parameters — requires careful optimization for mobile/edge deployment","Attention patterns are local (window-based) — may miss long-range dependencies critical for some scenes","Fine-tuning requires modifying the 150-class output head — cannot reuse pretrained class embeddings for new classes","Transfer learning effectiveness depends on domain similarity — fine-tuning on outdoor scenes (COCO) may require 50%+ of original training data","builder identity is not verified yet","no observed match outcomes yet"],"rank_breakdown":{"adoption":0.5156297097732508,"quality":0.5,"ecosystem":0.5000000000000001,"match_graph":0.25,"freshness":0.75,"weights":{"adoption":0.35,"quality":0.2,"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":"2026-05-24T12:16:22.766Z","last_scraped_at":"2026-05-03T14:23:00.162Z","last_commit":null},"community":{"stars":null,"forks":null,"weekly_downloads":null,"model_downloads":90906,"model_likes":35}},"distribution":{"claim_url":"https://unfragile.ai/submit?claim=shi-labs--oneformer_ade20k_swin_large","compare_url":"https://unfragile.ai/compare?artifact=shi-labs--oneformer_ade20k_swin_large"}},"signature":"fvvEN8yPsAvGFv41VamgbnuJPm6oRyg1PvG7rcK2bFIzAdItRTmmcjx+u4b7Cd6gAN/CekwyMWDURTZErtliAw==","signedAt":"2026-06-22T07:18:09.324Z","signedBy":"unfragile.ai","version":1},"_links":{"self":"https://unfragile.ai/api/v1/passport/shi-labs--oneformer_ade20k_swin_large","artifact":"https://unfragile.ai/shi-labs--oneformer_ade20k_swin_large","verify":"https://unfragile.ai/api/v1/verify?slug=shi-labs--oneformer_ade20k_swin_large","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"}}