{"passport":{"unfragile":{"@version":"1.0","version":"2026-05","artifact":{"id":"awesome-segment-anything-sam","slug":"segment-anything-sam","name":"Segment Anything (SAM)","type":"model","url":"https://arxiv.org/abs/2304.02643","page_url":"https://unfragile.ai/segment-anything-sam","categories":["productivity"],"tags":[],"pricing":{"model":"unknown","free":false,"starting_price":null},"status":"inactive","verified":false},"capabilities":[{"id":"awesome-segment-anything-sam__cap_0","uri":"capability://image.visual.promptable.image.segmentation.with.point.and.box.inputs","name":"promptable image segmentation with point and box inputs","description":"Segment Anything uses a vision transformer encoder-decoder architecture that accepts flexible prompts (points, bounding boxes, text, or masks) to segment any object in an image without task-specific fine-tuning. The model encodes the image once with a ViT backbone, then uses a lightweight mask decoder that processes prompt embeddings to generate segmentation masks in real-time. This prompt-based approach enables zero-shot segmentation across diverse object categories without retraining.","intents":["segment arbitrary objects in images by clicking points or drawing boxes without pre-defining object classes","build interactive annotation tools that respond to user prompts in real-time","extract object masks from images for downstream computer vision tasks without task-specific model training","enable non-experts to perform precise image segmentation through intuitive point-and-click interfaces"],"best_for":["computer vision engineers building interactive annotation platforms","teams automating image preprocessing pipelines for object detection or instance segmentation","researchers prototyping segmentation-dependent applications without labeled training data","product teams building image editing or content moderation tools requiring precise object isolation"],"limitations":["requires full image encoding pass for each inference, adding ~500ms latency on CPU for high-resolution images","prompt ambiguity can produce multiple valid segmentations; model returns single mask without ranking alternatives","performance degrades on small objects (<5% image area) and heavily occluded instances","no built-in temporal consistency for video segmentation; requires external frame-to-frame tracking","mask decoder assumes single-object focus per prompt; multi-object segmentation requires sequential prompting"],"requires":["PyTorch 1.9+","CUDA 11.0+ for GPU inference (CPU inference possible but slow)","minimum 4GB VRAM for batch processing","image input resolution typically 1024x1024 or compatible aspect ratios"],"input_types":["image (RGB, PNG/JPG/TIFF formats)","point prompts (x,y coordinates with foreground/background labels)","bounding box prompts (x1,y1,x2,y2 format)","mask prompts (binary masks as reference)","text prompts (optional, via CLIP integration)"],"output_types":["binary segmentation mask (H×W boolean array)","confidence scores per mask","bounding box of segmented region","polygon coordinates for mask boundary"],"categories":["image-visual","interactive-segmentation"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"awesome-segment-anything-sam__cap_1","uri":"capability://image.visual.automatic.mask.generation.for.full.image.segmentation","name":"automatic mask generation for full image segmentation","description":"SAM includes an automatic mask generation mode that systematically grids the image with point prompts and runs the segmentation decoder on each grid cell to produce a comprehensive set of non-overlapping masks covering all salient objects. The system uses non-maximum suppression and confidence filtering to deduplicate overlapping masks and retain only high-quality segmentations. This enables one-shot full-image instance segmentation without manual prompting.","intents":["automatically segment all objects in an image for instance segmentation without per-object prompting","generate training data for downstream object detection or segmentation models","create comprehensive object inventories for image understanding or asset management systems","preprocess images for batch segmentation pipelines without interactive user input"],"best_for":["data annotation teams automating mask generation for training datasets","computer vision pipelines requiring full-image instance segmentation at scale","content management systems needing automatic object extraction from image libraries","researchers evaluating segmentation quality across diverse image domains"],"limitations":["grid-based prompting is computationally expensive; full-image generation takes 30-60 seconds on GPU for 1024x1024 images","produces redundant overlapping masks that require post-processing to convert to non-overlapping instance segmentation","struggles with small objects and thin structures due to grid resolution constraints","no semantic understanding; cannot distinguish between object categories or filter by class","memory usage scales quadratically with image resolution due to mask storage"],"requires":["PyTorch 1.9+","CUDA 11.0+ for practical performance (CPU inference prohibitively slow)","minimum 8GB VRAM for batch processing multiple images","image resolution typically 1024x1024; larger images require tiling strategies"],"input_types":["image (RGB, PNG/JPG/TIFF formats)","grid resolution parameter (default 64x64 points)","confidence threshold for mask filtering"],"output_types":["list of binary segmentation masks (variable count per image)","confidence scores per mask","bounding boxes for each mask","area and stability metrics for filtering"],"categories":["image-visual","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"awesome-segment-anything-sam__cap_2","uri":"capability://image.visual.vision.transformer.image.encoding.with.hierarchical.feature.extraction","name":"vision transformer image encoding with hierarchical feature extraction","description":"SAM uses a Vision Transformer (ViT) backbone to encode images into dense feature maps that capture multi-scale visual information. The encoder processes the full image at once, producing hierarchical feature representations that preserve spatial structure while enabling the lightweight decoder to generate masks from arbitrary prompts. This design choice enables efficient amortization of computation across multiple prompts on the same image.","intents":["extract rich visual features from images for downstream segmentation and analysis tasks","enable efficient multi-prompt inference by reusing a single image encoding","capture both local detail and global context for accurate object boundary detection","support transfer learning by leveraging pre-trained ViT weights from large-scale vision datasets"],"best_for":["computer vision engineers building systems requiring efficient multi-prompt inference","researchers studying vision transformer architectures for dense prediction tasks","teams deploying segmentation models on resource-constrained devices via feature caching","developers integrating SAM into larger vision pipelines where feature reuse is critical"],"limitations":["ViT encoding adds ~300-500ms latency per image on CPU; GPU required for real-time performance","requires fixed input resolution (typically 1024x1024); aspect ratio changes require padding or resizing","memory footprint of encoded features scales with image resolution; high-res images require tiling","ViT backbone is computationally expensive; cannot be easily replaced with lighter models without retraining","no built-in support for temporal consistency across video frames; requires external tracking"],"requires":["PyTorch 1.9+","CUDA 11.0+ for GPU acceleration","minimum 4GB VRAM for single-image encoding","pre-trained ViT weights (provided with SAM model checkpoint)"],"input_types":["image (RGB, PNG/JPG/TIFF formats)","image resolution (default 1024x1024, supports variable aspect ratios with padding)"],"output_types":["dense feature maps (C×H×W tensors)","hierarchical feature representations at multiple scales","positional embeddings for spatial localization"],"categories":["image-visual","data-processing-analysis"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"awesome-segment-anything-sam__cap_3","uri":"capability://image.visual.lightweight.mask.decoder.with.prompt.embedding.fusion","name":"lightweight mask decoder with prompt embedding fusion","description":"SAM's mask decoder is a small transformer-based module that fuses image features from the ViT encoder with prompt embeddings (points, boxes, or masks) to generate segmentation masks. The decoder uses cross-attention mechanisms to align prompt information with image features, producing binary masks and confidence scores in real-time. This lightweight design enables fast inference and enables the decoder to be trained independently from the frozen image encoder.","intents":["generate segmentation masks from diverse prompt types (points, boxes, masks) in real-time","fuse spatial prompt information with global image context for accurate object boundaries","enable efficient training of the segmentation module without retraining the image encoder","support ambiguity resolution by generating multiple mask candidates for a single prompt"],"best_for":["interactive segmentation applications requiring sub-100ms mask generation latency","mobile or edge deployment scenarios where model size and inference speed are critical","researchers studying prompt-based dense prediction and attention mechanisms","teams fine-tuning SAM for domain-specific segmentation tasks"],"limitations":["decoder produces single mask per prompt; ambiguous prompts may require iterative refinement","cross-attention mechanism adds ~50-100ms latency per prompt on CPU","no built-in support for multi-class segmentation; requires external classification post-processing","prompt embedding quality depends on encoder; errors in feature extraction propagate to mask generation","decoder training requires large-scale mask annotations; limited labeled data reduces fine-tuning effectiveness"],"requires":["PyTorch 1.9+","pre-computed image features from ViT encoder","prompt embeddings (generated from point/box/mask inputs)","minimum 2GB VRAM for inference"],"input_types":["image features (C×H×W tensors from ViT encoder)","point prompts (x,y coordinates with foreground/background labels)","bounding box prompts (x1,y1,x2,y2 format)","mask prompts (binary masks as reference)"],"output_types":["binary segmentation mask (H×W boolean array)","confidence score (0-1 scalar)","multiple mask candidates (optional, for ambiguous prompts)"],"categories":["image-visual","tool-use-integration"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"awesome-segment-anything-sam__cap_4","uri":"capability://image.visual.ambiguity.aware.mask.generation.with.multiple.candidate.outputs","name":"ambiguity-aware mask generation with multiple candidate outputs","description":"SAM's decoder can generate multiple mask candidates for ambiguous prompts (e.g., a point on an object boundary could belong to multiple objects). The model produces a primary mask plus one or more alternative masks with associated confidence scores, enabling downstream systems to rank or select the most appropriate segmentation. This design acknowledges that segmentation is inherently ambiguous and provides tools for disambiguation.","intents":["handle ambiguous prompts by generating multiple plausible segmentations for user selection","enable interactive refinement where users can choose between candidate masks or provide additional prompts","quantify segmentation uncertainty for downstream decision-making in automated pipelines","support applications requiring multiple valid interpretations of object boundaries"],"best_for":["interactive annotation tools where users disambiguate segmentation results","uncertainty-aware computer vision pipelines that need confidence estimates","applications requiring human-in-the-loop refinement of segmentation results","research systems studying segmentation ambiguity and multi-modal outputs"],"limitations":["multiple mask generation increases inference latency by 20-30% compared to single-mask mode","ranking multiple masks requires external criteria (user preference, downstream task loss, etc.)","ambiguity detection is implicit; no explicit mechanism to signal when ambiguity is high","memory usage increases linearly with number of candidate masks","no semantic understanding of why masks differ; cannot explain ambiguity sources to users"],"requires":["PyTorch 1.9+","pre-computed image features from ViT encoder","prompt embeddings (point, box, or mask format)","minimum 2GB VRAM for inference"],"input_types":["image features (C×H×W tensors from ViT encoder)","point prompts (x,y coordinates with foreground/background labels)","bounding box prompts (x1,y1,x2,y2 format)","mask prompts (binary masks as reference)"],"output_types":["primary binary segmentation mask (H×W boolean array)","alternative mask candidates (variable count, typically 1-3)","confidence scores per mask (0-1 scalars)","stability metrics indicating ambiguity level"],"categories":["image-visual","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"awesome-segment-anything-sam__cap_5","uri":"capability://data.processing.analysis.large.scale.mask.dataset.generation.and.curation.sa.1b","name":"large-scale mask dataset generation and curation (sa-1b)","description":"SAM was trained on SA-1B, a dataset of 1.1 billion segmentation masks automatically generated from 11 million images using an iterative process: initial SAM predictions were refined with human feedback, then used to generate additional masks via automatic prompting. This dataset construction process demonstrates how to bootstrap large-scale segmentation annotations without manual labeling, enabling SAM's zero-shot generalization across diverse object categories and image domains.","intents":["understand how to construct large-scale segmentation datasets through semi-automatic annotation","leverage SAM's training methodology to generate domain-specific mask datasets for fine-tuning","evaluate segmentation model generalization across diverse image domains and object categories","build annotation pipelines that combine automatic prediction with human feedback for quality control"],"best_for":["data engineering teams building large-scale annotation pipelines","researchers studying dataset construction and annotation quality","organizations fine-tuning SAM on domain-specific data (medical imaging, satellite imagery, etc.)","teams evaluating the impact of dataset diversity on model generalization"],"limitations":["SA-1B dataset is not publicly available; researchers cannot directly access or analyze the full dataset","iterative annotation process is computationally expensive; requires significant infrastructure for large-scale deployment","human feedback quality depends on annotator expertise; domain-specific feedback may be needed for specialized tasks","dataset bias towards natural images; limited coverage of specialized domains (medical, industrial, etc.)","no fine-grained category labels; masks are category-agnostic, limiting semantic understanding"],"requires":["access to large image corpus (11M+ images for comparable scale)","computational infrastructure for iterative SAM inference and mask generation","human annotators for quality control and feedback (optional but recommended)","storage capacity for 1B+ mask annotations (~100TB+ for full dataset)"],"input_types":["image corpus (PNG/JPG/TIFF formats)","initial SAM predictions (binary masks)","human feedback (mask refinements, quality labels)"],"output_types":["segmentation masks (binary, H×W arrays)","mask metadata (area, stability, quality scores)","image-mask associations for training dataset construction"],"categories":["data-processing-analysis","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"awesome-segment-anything-sam__cap_6","uri":"capability://image.visual.cross.domain.generalization.through.vision.transformer.pre.training","name":"cross-domain generalization through vision transformer pre-training","description":"SAM achieves zero-shot generalization across diverse image domains (natural images, medical imaging, satellite imagery, etc.) by leveraging a ViT encoder pre-trained on large-scale vision datasets. The encoder learns domain-agnostic visual features that transfer effectively to new domains without fine-tuning, while the lightweight mask decoder is trained on diverse segmentation masks from SA-1B. This design enables SAM to segment objects in domains not seen during training.","intents":["segment objects in specialized image domains (medical, satellite, microscopy) without domain-specific training","evaluate model generalization across diverse visual domains and object categories","build segmentation systems that adapt to new domains through fine-tuning rather than retraining from scratch","understand how pre-training and dataset diversity contribute to zero-shot generalization"],"best_for":["computer vision teams deploying segmentation systems across multiple image domains","researchers studying transfer learning and domain generalization in vision models","organizations building domain-agnostic segmentation tools (medical imaging, satellite analysis, etc.)","developers fine-tuning SAM for specialized domains with limited labeled data"],"limitations":["generalization degrades on highly specialized domains (e.g., microscopy, thermal imaging) without fine-tuning","ViT pre-training is computationally expensive; cannot easily replace with lighter models without retraining","domain shift can cause prompt ambiguity to increase; automatic mask generation may produce spurious results","no explicit mechanism to detect domain shift or estimate generalization confidence","fine-tuning on domain-specific data requires labeled masks; limited data reduces effectiveness"],"requires":["PyTorch 1.9+","pre-trained SAM checkpoint (ViT encoder + mask decoder)","CUDA 11.0+ for GPU inference (CPU inference possible but slow)","minimum 4GB VRAM for inference"],"input_types":["image (RGB, PNG/JPG/TIFF formats from any domain)","point prompts (x,y coordinates with foreground/background labels)","bounding box prompts (x1,y1,x2,y2 format)","mask prompts (binary masks as reference)"],"output_types":["binary segmentation mask (H×W boolean array)","confidence score (0-1 scalar)","multiple mask candidates (optional, for ambiguous prompts)"],"categories":["image-visual","memory-knowledge"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"awesome-segment-anything-sam__cap_7","uri":"capability://image.visual.fine.tuning.and.adaptation.for.domain.specific.segmentation","name":"fine-tuning and adaptation for domain-specific segmentation","description":"SAM can be fine-tuned on domain-specific segmentation data by training the lightweight mask decoder on labeled masks from the target domain while keeping the ViT encoder frozen. This approach enables rapid adaptation to specialized domains (medical imaging, satellite imagery, etc.) with limited labeled data, reducing fine-tuning time and data requirements compared to training end-to-end models. The frozen encoder preserves domain-agnostic visual features while the decoder learns domain-specific segmentation patterns.","intents":["adapt SAM to specialized image domains (medical, satellite, microscopy) with limited labeled data","fine-tune SAM on domain-specific datasets to improve segmentation accuracy without retraining from scratch","evaluate the effectiveness of transfer learning for segmentation across diverse domains","build domain-specific segmentation systems that leverage SAM's pre-trained features"],"best_for":["organizations deploying segmentation systems in specialized domains (medical, satellite, industrial)","researchers studying transfer learning and few-shot learning for dense prediction tasks","teams with limited labeled data that need to adapt SAM to new domains","developers building domain-specific annotation tools or preprocessing pipelines"],"limitations":["fine-tuning requires labeled segmentation masks; limited data reduces effectiveness and risks overfitting","frozen encoder may not capture domain-specific visual features (e.g., medical imaging artifacts)","fine-tuning time depends on dataset size; large datasets require significant computational resources","no built-in mechanisms for detecting when fine-tuning has converged or when domain shift is too large","fine-tuned models may lose zero-shot generalization to other domains; requires careful regularization"],"requires":["PyTorch 1.9+","pre-trained SAM checkpoint (ViT encoder + mask decoder)","labeled segmentation masks for target domain (100+ masks recommended)","CUDA 11.0+ for GPU training (CPU training prohibitively slow)","minimum 8GB VRAM for training"],"input_types":["image (RGB, PNG/JPG/TIFF formats from target domain)","segmentation masks (binary, H×W arrays)","point prompts (x,y coordinates with foreground/background labels)","bounding box prompts (x1,y1,x2,y2 format)"],"output_types":["fine-tuned mask decoder checkpoint","segmentation masks (binary, H×W arrays)","confidence scores per mask","training metrics (loss, IoU, etc.)"],"categories":["image-visual","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"awesome-segment-anything-sam__cap_8","uri":"capability://image.visual.interactive.refinement.with.iterative.prompting","name":"interactive refinement with iterative prompting","description":"SAM supports interactive refinement workflows where users provide initial prompts (points or boxes), review the generated masks, and iteratively refine prompts to correct segmentation errors. The system reuses the frozen image encoding across refinement iterations, enabling sub-100ms mask generation for each refinement step. This design enables efficient human-in-the-loop annotation where users guide the model toward correct segmentations through iterative feedback.","intents":["enable interactive annotation tools where users refine segmentation results through iterative prompting","build annotation pipelines that combine automatic prediction with human feedback for quality control","support real-time segmentation refinement in interactive applications (image editing, content moderation)","quantify annotation effort by measuring the number of prompts required to achieve target accuracy"],"best_for":["annotation teams using interactive tools to label segmentation datasets","interactive image editing applications requiring precise object isolation","content moderation systems where human reviewers refine automatic segmentations","research systems studying human-in-the-loop machine learning for dense prediction"],"limitations":["iterative refinement requires user interaction; cannot be fully automated for ambiguous cases","refinement effectiveness depends on user expertise; non-experts may struggle with complex objects","no built-in guidance for users on how to refine prompts; requires domain knowledge","refinement time depends on object complexity; complex objects may require many iterations","no mechanism to estimate when refinement has converged or when further iterations are unlikely to help"],"requires":["PyTorch 1.9+","pre-computed image features from ViT encoder","interactive interface for prompt input (web app, desktop tool, etc.)","minimum 2GB VRAM for inference","sub-100ms latency requirement for interactive responsiveness"],"input_types":["image (RGB, PNG/JPG/TIFF formats)","point prompts (x,y coordinates with foreground/background labels)","bounding box prompts (x1,y1,x2,y2 format)","mask prompts (binary masks as reference)","user feedback (mask corrections, refinement guidance)"],"output_types":["binary segmentation mask (H×W boolean array)","confidence score (0-1 scalar)","refinement history (sequence of prompts and masks)","annotation metadata (time, number of iterations, etc.)"],"categories":["image-visual","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"awesome-segment-anything-sam__cap_9","uri":"capability://image.visual.efficient.inference.with.model.quantization.and.optimization","name":"efficient inference with model quantization and optimization","description":"SAM supports various inference optimizations including model quantization (INT8, FP16), knowledge distillation to smaller models, and hardware-specific optimizations (ONNX, TensorRT) to enable deployment on resource-constrained devices. These optimizations reduce model size by 4-8x and inference latency by 2-4x while maintaining segmentation quality, enabling SAM deployment on mobile devices, edge hardware, and real-time applications. The frozen encoder design facilitates efficient optimization by decoupling image encoding from mask generation.","intents":["deploy SAM on mobile devices and edge hardware with limited computational resources","optimize SAM inference for real-time applications requiring sub-100ms latency","reduce model size for efficient storage and distribution across edge devices","evaluate the trade-offs between model size, inference latency, and segmentation accuracy"],"best_for":["mobile and edge deployment teams requiring efficient segmentation inference","real-time applications (video processing, robotics) with strict latency budgets","organizations deploying segmentation systems at scale with limited computational resources","researchers studying model compression and efficient inference for dense prediction"],"limitations":["quantization can reduce segmentation accuracy by 1-5% depending on quantization scheme","knowledge distillation requires training smaller student models; adds development overhead","hardware-specific optimizations (ONNX, TensorRT) require platform-specific tuning","optimization effectiveness varies across hardware platforms; requires benchmarking on target devices","optimized models may lose compatibility with certain prompt types or inference modes"],"requires":["PyTorch 1.9+ or ONNX Runtime 1.12+","quantization tools (PyTorch quantization, TensorRT, etc.)","target hardware specifications (mobile device, edge device, etc.)","benchmark datasets for evaluating optimization trade-offs"],"input_types":["pre-trained SAM checkpoint (ViT encoder + mask decoder)","quantization configuration (bit-width, scheme, etc.)","hardware specifications (device type, memory constraints, etc.)","benchmark dataset for accuracy evaluation"],"output_types":["optimized model checkpoint (quantized, distilled, etc.)","inference latency benchmarks (ms per image)","model size metrics (MB, parameter count)","accuracy metrics (IoU, F1, etc.) on benchmark dataset"],"categories":["image-visual","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0}],"trust":{"score":21,"verified":false,"data_access_risk":"low","permissions":["PyTorch 1.9+","CUDA 11.0+ for GPU inference (CPU inference possible but slow)","minimum 4GB VRAM for batch processing","image input resolution typically 1024x1024 or compatible aspect ratios","CUDA 11.0+ for practical performance (CPU inference prohibitively slow)","minimum 8GB VRAM for batch processing multiple images","image resolution typically 1024x1024; larger images require tiling strategies","CUDA 11.0+ for GPU acceleration","minimum 4GB VRAM for single-image encoding","pre-trained ViT weights (provided with SAM model checkpoint)"],"failure_modes":["requires full image encoding pass for each inference, adding ~500ms latency on CPU for high-resolution images","prompt ambiguity can produce multiple valid segmentations; model returns single mask without ranking alternatives","performance degrades on small objects (<5% image area) and heavily occluded instances","no built-in temporal consistency for video segmentation; requires external frame-to-frame tracking","mask decoder assumes single-object focus per prompt; multi-object segmentation requires sequential prompting","grid-based prompting is computationally expensive; full-image generation takes 30-60 seconds on GPU for 1024x1024 images","produces redundant overlapping masks that require post-processing to convert to non-overlapping instance segmentation","struggles with small objects and thin structures due to grid resolution constraints","no semantic understanding; cannot distinguish between object categories or filter by class","memory usage scales quadratically with image resolution due to mask storage","builder identity is not verified yet","no observed match outcomes yet"],"rank_breakdown":{"adoption":0.05,"quality":0.35,"ecosystem":0.25,"match_graph":0.25,"freshness":0.5,"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":"inactive","updated_at":"2026-06-17T09:51:04.049Z","last_scraped_at":"2026-05-03T14:00:27.894Z","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=segment-anything-sam","compare_url":"https://unfragile.ai/compare?artifact=segment-anything-sam"}},"signature":"kbK1Q4HyTlOy9eMaeG7EnILeSt2rjkftYgBCdsObwyJJX7uI40jJ7q5SMFyRB57ZIAbRU1m0ba3fMrGcvny6BQ==","signedAt":"2026-06-21T03:08:50.372Z","signedBy":"unfragile.ai","version":1},"_links":{"self":"https://unfragile.ai/api/v1/passport/segment-anything-sam","artifact":"https://unfragile.ai/segment-anything-sam","verify":"https://unfragile.ai/api/v1/verify?slug=segment-anything-sam","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"}}