.. _Model Catalog:


=============
Model Catalog
=============

This catalog contains **prebuilt AI models** optimized for **SiMa AI accelerators** and other embedded platforms.
These models are designed for **fast inference, efficient processing, and real-world deployment** across various AI
domains such as **image classification, object detection, pose estimation, depth estimation, and anomaly detection**.
The catalog is organized into **major AI categories**, helping developers quickly find the right models for their needs.

Image Classification Models
---------------------------

These models are optimized for **image classification**, helping identify objects, scenes, and patterns in images.

.. list-table:: Image Classification Models
   :header-rows: 1

   * - Model Name
     - Description
   * - alexnet
     - Classic CNN model for image classification
   * - densenet121_12
     - DenseNet-121 variant for classification
   * - densenet121_9
     - Optimized DenseNet-121 for edge AI
   * - densenet_121
     - Standard DenseNet-121 for deep feature extraction
   * - densenet_161
     - Deeper DenseNet for high-accuracy classification
   * - densenet_169
     - Balanced DenseNet variant for classification
   * - densenet_201
     - Large DenseNet model for advanced recognition
   * - efficientnet-lite4-11
     - EfficientNet-Lite for mobile and embedded devices
   * - efficientnet-v2-b0
     - EfficientNetV2 with optimized feature extraction
   * - efficientnet-v2-m
     - Medium variant of EfficientNetV2 for accuracy
   * - googlenet
     - Inception-based model for image classification
   * - regnet_x_1.6gf
     - RegNet optimized for low-power image tasks
   * - regnet_x_3.2gf
     - Scalable RegNet model with improved accuracy
   * - regnet_y_8gf
     - High-performance RegNet-Y variant
   * - resnet50
     - Standard ResNet-50 for robust classification
   * - resnet101_32x8d
     - Wider ResNet-101 for detailed feature extraction
   * - vgg16
     - Popular VGG model for classification
   * - vgg19_bn
     - Batch-normalized VGG19 for improved convergence
   * - wide-resnet-101
     - Wide ResNet for handling complex datasets

Object Detection Models
-----------------------

These models are designed for **real-time object detection**, tracking multiple objects in images and videos.

.. list-table:: Object Detection Models
   :header-rows: 1

   * - Model Name
     - Description
   * - centernet
     - CenterNet for efficient object detection
   * - yolov3
     - Classic YOLOv3 model for real-time detection
   * - yolov3_tiny
     - Lightweight YOLOv3 for edge devices
   * - yolov7
     - High-performance YOLO model with fast inference
   * - yolov7x
     - Extended YOLOv7 with improved accuracy
   * - yolov8l
     - Large variant of YOLOv8 for high-precision detection
   * - yolov8n
     - Nano YOLOv8 optimized for low-power devices
   * - yolov9c
     - Experimental YOLOv9 variant for advanced tasks
   * - yolox_s
     - Small YOLOX variant for edge deployment
   * - yolox_x
     - Extra-large YOLOX model for best performance

Pose Estimation Models
----------------------

These models are used for **human pose estimation**, detecting keypoints in body movement and gestures.

.. list-table:: Pose Estimation Models
   :header-rows: 1

   * - Model Name
     - Description
   * - open_pose
     - Human pose estimation for activity tracking
   * - hrnet_w32
     - High-resolution pose estimation model
   * - face_landmark
     - Facial keypoint detection for AR applications
   * - fld_68landmarks
     - 68-point facial landmark detection
   * - regnet_x_32gf
     - High-performance backbone for pose tasks

Depth Estimation Models
-----------------------

These models estimate **depth information** from single or stereo images, useful for **3D reconstruction and robotics**.

.. list-table:: Depth Estimation Models
   :header-rows: 1

   * - Model Name
     - Description
   * - depth_anything_v2_vits
     - General-purpose depth estimation model
   * - midas_v21_small_256
     - MiDaS depth estimation optimized for efficiency
   * - optimized_midas
     - High-speed depth estimation with accuracy balance
   * - lightstereo
     - Lightweight stereo depth estimation
   * - yolov7_gelan_c_seg
     - YOLO-based depth and segmentation model

Anomaly Detection & Feature Extraction
--------------------------------------

These models are used for **detecting anomalies**, identifying defects, and learning representations.

.. list-table:: Anomaly Detection & Feature Extraction
   :header-rows: 1

   * - Model Name
     - Description
   * - efficient_ad
     - Efficient Anomaly Detection for industrial use
   * - fastflow_demo
     - Fast anomaly detection for real-time applications
   * - fastflow_mvtec
     - MVTec-trained FastFlow model for defect detection
   * - mnist_cnn
     - CNN model trained on the MNIST dataset
   * - reid
     - Person re-identification model for surveillance
   * - stfpm_mvtec
     - Self-teaching feature pyramid anomaly detection