Timm

General Information

Field	Value
Package Name	timm
Manufacturer / Vendor	Ross Wightman / Hugging Face
Software Category	Library
Primary Documentation	Documentation, GitHub, PyPI
Programming Language(s)	Python
License	Apache License 2.0
Deployed Version(s)	>=1.0.9, >=1.0.14 (version-locked at 1.0.22 and 1.0.24)
Most Recent Available Version	1.0.24
Last Review Date	2026-01-26

Overview

timm (PyTorch Image Models) is a comprehensive deep learning library providing state-of-the-art computer vision models, layers, utilities, optimizers, schedulers, data loaders, and augmentation techniques. The library contains the largest collection of PyTorch image encoders and backbones, including architectures such as ResNet, EfficientNet, Vision Transformer (ViT), ConvNeXt, MobileNet, and many others.

Within the medical device software, timm serves as the foundational backbone provider for all image classification and feature extraction pipelines. The library enables transfer learning by providing pre-trained model architectures that can be adapted for dermatological image analysis tasks. The primary function used is timm.create_model(), which instantiates neural network backbones with configurable parameters.

timm was selected over alternatives due to:

• Comprehensive model coverage with 1,200+ pre-trained architectures
• Active maintenance and regular updates from Hugging Face
• Consistent API across different model families
• Extensive pre-training on ImageNet-1k and ImageNet-22k datasets
• Strong community adoption (36k+ GitHub stars) and documentation
• Apache 2.0 license permitting commercial use

Functional Requirements

The following functional capabilities of this SOUP are relied upon by the medical device software.

Requirement ID	Description	Source / Reference
FR-001	Create model instances from architecture name strings with configurable parameters	timm.create_model() function
FR-002	Provide ConvNeXt family backbones (ConvNeXtV2, ConvNeXt-Small) for condition and pattern classifiers	ConvNeXt model implementations
FR-003	Provide EfficientNet family backbones (B0, B5) for quality assessment and domain classification	EfficientNet model implementations
FR-004	Support extraction of backbone feature dimensions via model.num_features attribute	Model architecture API
FR-005	Allow removal of classification head via num_classes=0 parameter for custom head attachment	create_model() parameter handling
FR-006	Support loading of pre-trained ImageNet weights via pretrained=True parameter	Pre-trained weight loading mechanism
FR-007	Expose hidden layer dimensions via model.head_hidden_size for proper classifier head dimensioning	Model attribute access

Performance Requirements

The following performance expectations are relevant to the medical device software.

Requirement ID	Description	Acceptance Criteria
PR-001	Model instantiation completes within acceptable initialization time	Model creation < 30 seconds on target hardware
PR-002	Feature extraction maintains numerical precision for downstream processing	Float32 tensor output with standard IEEE 754 precision
PR-003	Memory footprint allows concurrent model loading for multi-task pipelines	Model fits within allocated GPU memory (varies by model size)

Hardware Requirements

The following hardware dependencies or constraints are imposed by this SOUP component.

Requirement ID	Description	Notes / Limitations
HR-001	CUDA-compatible GPU recommended for efficient inference	CPU inference supported but significantly slower
HR-002	Sufficient GPU memory for model architecture	EfficientNet-B5: ~30MB, ConvNeXtV2-Base: ~350MB
HR-003	System memory for model weight loading and tensor storage	Minimum 8GB RAM recommended for model operations

Software Requirements

The following software dependencies and environmental assumptions are required by this SOUP component.

Requirement ID	Description	Dependency / Version Constraints
SR-001	Python runtime environment	Python >=3.8
SR-002	PyTorch deep learning framework	torch >=2.0.0 (recommend >=2.6.0)
SR-003	torchvision for image transformations	torchvision >=0.15.0
SR-004	Hugging Face Hub for model weight downloads	huggingface_hub (optional, for weights)

Known Anomalies Assessment

This section evaluates publicly reported issues, defects, or security vulnerabilities associated with this SOUP component and their relevance to the medical device software.

Anomaly Reference	Status	Applicable	Rationale	Reviewed At
CVE-2025-32434 (PyTorch torch.load() RCE vulnerability)	Fixed	No	Affects PyTorch <=2.5.1; the device uses version-locked PyTorch dependencies. When loading timm models, weights_only=True combined with PyTorch >=2.6.0 mitigates this vulnerability. Production deployments use pre-verified, internally cached model weights	2026-01-26

The timm library is maintained by Hugging Face with active development (36k+ GitHub stars, 5k+ forks, regular releases). According to Snyk's security analysis, the package has been scanned and no known vulnerabilities have been identified. The project does not maintain a formal security policy (SECURITY.md), which is common for research-originated open-source projects, but vulnerabilities in the underlying PyTorch framework are addressed promptly through version updates.

The device's usage pattern minimizes attack surface exposure:

• No dynamic model loading from external sources: All model architectures are instantiated via timm.create_model() with predefined architecture names; no arbitrary model files are loaded from user input
• Pre-trained weights from verified sources: Model weights are sourced exclusively from the Hugging Face model hub and cached internally before deployment
• Version locking: Requirements lock files (requirements_lock.txt) pin timm and PyTorch to verified versions, ensuring reproducible and auditable deployments
• Limited API surface: The device uses only create_model() for backbone instantiation and num_features/head_hidden_size attributes for dimensioning; no experimental or unsafe features are utilized

Risk Control Measures

The following risk control measures are implemented to mitigate potential security and operational risks associated with this SOUP component:

• Version locking via requirements_lock.txt files ensures reproducible deployments
• Pre-trained weights are sourced from verified Hugging Face model hub
• Model weights are cached and verified before deployment
• No dynamic model loading from external sources; all architectures are predefined
• Limited API surface usage (create_model() and attribute access only)

Assessment Methodology

Known anomalies were identified and assessed using the following methodology:

• Sources consulted:

  • National Vulnerability Database (NVD) search for "timm" and "pytorch-image-models"
  • GitHub Security Advisories for the huggingface/pytorch-image-models repository
  • PyPI package security reports
  • Dependency vulnerability scanners (pip-audit, safety)
  • PyTorch framework security advisories (as timm's primary dependency)

• Criteria for determining applicability:

  • Vulnerability must affect deployed versions (1.0.9+)
  • Vulnerability must be exploitable in the device's operational context
  • Vulnerability must impact the specific timm functions used (primarily create_model())

Signature meaning

The signatures for the approval process of this document can be found in the verified commits at the repository for the QMS. As a reference, the team members who are expected to participate in this document and their roles in the approval process, as defined in Annex I Responsibility Matrix of the GP-001, are:

• Author: Team members involved
• Reviewer: JD-003 Design & Development Manager, JD-004 Quality Manager & PRRC
• Approver: JD-001 General Manager