rustytorch/crates/models/rtx-vision/src/lib.rs

//! # RTX Vision - Vision Transformers and Computer Vision Models
//!
//! Revolutionary vision models with world-class performance for RustyTorch++.
//!
//! ## Features
//! - Vision Transformer (ViT) with Flash Attention integration
//! - ConvNeXt modern ConvNet architecture
//! - EfficientNet V2 with Fused-MBConv blocks and progressive learning
//! - MobileViT hybrid CNN-Transformer for mobile deployment
//! - EdgeViT edge-optimized Vision Transformer for efficient deployment
//! - Mobile-optimized components (depthwise separable convolutions, efficient attention)
//! - CLIP vision encoders for multimodal learning
//! - Patch embedding with learnable positional encodings
//! - Advanced image preprocessing and augmentation
//! - Quantization-aware training utilities
//! - PyTorch-compatible APIs where applicable
//!
//! ## Architecture
//! - `preprocessing`: Image loading, normalization, and augmentation
//! - `layers`: Patch embedding, positional encoding, specialized layers
//! - `models`: Complete vision model implementations
//! - `utils`: Vision-specific utilities and metrics
//!
//! ## Example
//! ```rust,ignore
//! use rtx_vision::{models::ViT, preprocessing::ImageTensor};
//!
//! let model = ViT::new(ViTConfig::base_16()).unwrap();
//! let image = ImageTensor::from_path("image.jpg").unwrap();
//! let output = model.forward(&image).unwrap();
//! ```

pub mod architectures;
pub mod error;
pub mod layers;
pub mod models;
pub mod preprocessing;
pub mod utils;

// Vision tensor implementation (using real rtx-tensor)
pub mod vision_tensor;

#[cfg(test)]
pub mod vision_tensor_tests;

// Re-export main types
pub use error::{Result, VisionError};

// Re-export tensor types
pub use rtx_tensor::{DType, Device, Result as TensorResult, Shape, Tensor, TensorError};

// Re-export preprocessing types
pub use preprocessing::{Augmentation, ImageProcessor, ImageTensor};

// Re-export layer types
pub use layers::{PatchEmbedding, PositionalEncoding};

// Re-export model types
pub use models::{ConvNeXt, ConvNeXtConfig, ViT, ViTConfig};

// Re-export architecture types
pub use architectures::{
    AGCConfig,
    AdaptiveGradientClipping,
    BasicBlock,
    BlockAttention,
    BlockAttentionConfig,
    Bottleneck,
    // CoAtNet
    CoAtNet,
    CoAtNetConfig,
    CoAtNetMBConvBlock,
    CoAtNetMBConvConfig,
    CoAtNetStage,
    CoAtNetTransformerBlock,
    CoAtNetTransformerConfig,
    CoAtNetVariant,
    // ConvNeXt V2
    ConvNeXtV2,
    ConvNeXtV2Block,
    ConvNeXtV2Config,
    DecomposedAttention,
    DecomposedAttentionConfig,
    DenseBlock,
    DenseLayer,
    // DenseNet
    DenseNet,
    DenseNetConfig,
    DenseNetVariant,
    DepthwiseSepConv,
    DepthwiseSepConvConfig,
    // Mobile Components
    DepthwiseSeparableConv,
    DepthwiseSeparableConvConfig,
    // EdgeViT
    EdgeViT,
    EdgeViTClassificationHead,
    EdgeViTConfig,
    EdgeViTVariant,
    EfficientAttention,
    EfficientAttentionConfig,
    EfficientChannelAttention,
    EfficientChannelAttentionConfig,
    // EfficientNet V2
    EfficientNetV2,
    EfficientNetV2Config,
    EfficientNetV2Variant,
    EfficientPositionalEncoding,
    FusedMBConvBlock,
    FusedMBConvConfig,
    GlobalResponseNormalization,
    GridAttention,
    GridAttentionConfig,
    InvertedResidualBlock,
    InvertedResidualConfig,
    LGLBlock,
    LGLBlockConfig,
    MAEDecoder,
    MAEEncoder,
    // MaxViT
    MaxViT,
    MaxViTConfig,
    MaxViTMBConvBlock,
    MaxViTMBConvConfig,
    MaxViTStage,
    MaxViTVariant,
    MobileActivations,
    // MobileNet
    MobileNet,
    MobileNetConfig,
    MobileNetVariant,
    // MobileViT
    MobileViT,
    MobileViTBlock,
    MobileViTBlockConfig,
    MobileViTConfig,
    MobileViTTransformerBlock,
    MobileViTVariant,
    MultiAxisAttention,
    MultiAxisAttentionConfig,
    MultiScaleAggregation,
    MultiScaleAggregationConfig,
    NFBlock,
    NFBlockConfig,
    // NFNet
    NFNet,
    NFNetConfig,
    NFNetHead,
    NFNetStem,
    NFNetTransition,
    NFNetVariant,
    QuantizationAwareUtils,
    // RegNet
    RegNet,
    RegNetBlock,
    RegNetBlockConfig,
    RegNetConfig,
    RegNetHead,
    RegNetStage,
    RegNetStageConfig,
    RegNetStem,
    RelativeAttention,
    RelativeAttentionConfig,
    // ResNet
    ResNet,
    ResNetConfig,
    ResNetVariant,
    SEModule,
    SEModule as RegNetSEModule,
    SEModuleConfig,
    SEModuleConfig as RegNetSEModuleConfig,
    ScaledWSConfig,
    ScaledWeightStandardization,
    TransformerBlockConfig,
    // VGG
    VGG,
    VGGConfig,
    VGGVariant,
};

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_module_structure() {
        // Basic smoke test to ensure module structure is valid
        assert!(true);
    }
}