QwenVL model implementing all task interfaces with custom prompts

VisionReasoner model implementing all task interfaces

MLP as used in Vision Transformer, MLP-Mixer and related networks

VisionReasoner model implementing all task interfaces

DPT Head for dense prediction tasks. This implementation follows the architecture described in "Vision Transformers for Dense Prediction"

QwenVLCoT model implementing all task interfaces with custom prompts

ResidualBlock: construct a block of two conv layers with residual connections

Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).

MLP as used in Vision Transformer, MLP-Mixer and related networks

Residual convolution module.

The Aggregator applies alternating-attention over input frames, as described in VGGT: Visual Geometry Grounded Transformer. Remember to

CameraHead predicts camera parameters from token representations using iterative refinement. It applies a series of transformer blocks (the

Transformer model that updates track estimates.

Transformer model that updates track estimates.

Feature fusion block.

2D image to patch embedding: (B,C,H,W) -> (B,N,D) Args: img_size: Image size. patch_size: Patch token size. in_chans

Generates and caches 2D spatial positions for patches in a grid. This class efficiently manages the generation of spatial coordinates for patches

2D Rotary Position Embedding implementation. This module applies rotary position embeddings to input tokens based on their 2D spatial positio

Track head that uses DPT head to process tokens and BaseTrackerPredictor for tracking. The tracking is performed iteratively, refining predic

Abstract base class for vision models that process images and instructions

Interface for counting tasks

Interface for object detection tasks

Interface for visual question answering tasks

ResidualBlock: construct a block of two conv layers with residual connections

Interface for segmentation tasks