Add Ultralytics ViT Docs (#3230)

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

docs/hub/datasets.md

@@ -1,6 +1,6 @@
 ---
 comments: true
-description: Upload custom datasets to Ultralytics HUB for YOLOv5 and YOLOv8 models. Follow YAML structure, zip and upload. Scan & train new models.
+description: Efficiently manage and use custom datasets on Ultralytics HUB for streamlined training with YOLOv5 and YOLOv8 models.
 keywords: Ultralytics, HUB, Datasets, Upload, Visualize, Train, Custom Data, YAML, YOLOv5, YOLOv8
 ---
 

docs/reference/vit/rtdetr/model.md

@@ -0,0 +1,9 @@
+---
+description: Learn about the RTDETR model in Ultralytics YOLO Docs and how it can be used for object detection with improved speed and accuracy. Find implementation details and more.
+keywords: RTDETR, Ultralytics, YOLO, object detection, speed, accuracy, implementation details
+---
+
+## RTDETR
+---
+### ::: ultralytics.vit.rtdetr.model.RTDETR
+<br><br>

docs/reference/vit/rtdetr/predict.md

@@ -0,0 +1,9 @@
+---
+description: Learn about the RTDETRPredictor class and how to use it for vision transformer object detection with Ultralytics YOLO.
+keywords: RTDETRPredictor, object detection, vision transformer, Ultralytics YOLO
+---
+
+## RTDETRPredictor
+---
+### ::: ultralytics.vit.rtdetr.predict.RTDETRPredictor
+<br><br>

docs/reference/vit/rtdetr/train.md

@@ -0,0 +1,14 @@
+---
+description: Learn how to use RTDETRTrainer from Ultralytics YOLO Docs. Train object detection models with the latest VIT-based RTDETR system.
+keywords: RTDETRTrainer, Ultralytics YOLO Docs, object detection, VIT-based RTDETR system, train
+---
+
+## RTDETRTrainer
+---
+### ::: ultralytics.vit.rtdetr.train.RTDETRTrainer
+<br><br>
+
+## train
+---
+### ::: ultralytics.vit.rtdetr.train.train
+<br><br>

docs/reference/vit/rtdetr/val.md

@@ -0,0 +1,14 @@
+---
+description: Documentation for RTDETRValidator data validation tool in Ultralytics RTDETRDataset.
+keywords: RTDETRDataset, RTDETRValidator, data validation, documentation
+---
+
+## RTDETRDataset
+---
+### ::: ultralytics.vit.rtdetr.val.RTDETRDataset
+<br><br>
+
+## RTDETRValidator
+---
+### ::: ultralytics.vit.rtdetr.val.RTDETRValidator
+<br><br>

docs/reference/vit/sam/amg.md

@@ -0,0 +1,89 @@
+---
+description: Explore and learn about functions in Ultralytics MaskData library such as mask_to_rle_pytorch, area_from_rle, generate_crop_boxes, and more.
+keywords: Ultralytics, SAM, MaskData, mask_to_rle_pytorch, area_from_rle, generate_crop_boxes, batched_mask_to_box, documentation
+---
+
+## MaskData
+---
+### ::: ultralytics.vit.sam.amg.MaskData
+<br><br>
+
+## is_box_near_crop_edge
+---
+### ::: ultralytics.vit.sam.amg.is_box_near_crop_edge
+<br><br>
+
+## box_xyxy_to_xywh
+---
+### ::: ultralytics.vit.sam.amg.box_xyxy_to_xywh
+<br><br>
+
+## batch_iterator
+---
+### ::: ultralytics.vit.sam.amg.batch_iterator
+<br><br>
+
+## mask_to_rle_pytorch
+---
+### ::: ultralytics.vit.sam.amg.mask_to_rle_pytorch
+<br><br>
+
+## rle_to_mask
+---
+### ::: ultralytics.vit.sam.amg.rle_to_mask
+<br><br>
+
+## area_from_rle
+---
+### ::: ultralytics.vit.sam.amg.area_from_rle
+<br><br>
+
+## calculate_stability_score
+---
+### ::: ultralytics.vit.sam.amg.calculate_stability_score
+<br><br>
+
+## build_point_grid
+---
+### ::: ultralytics.vit.sam.amg.build_point_grid
+<br><br>
+
+## build_all_layer_point_grids
+---
+### ::: ultralytics.vit.sam.amg.build_all_layer_point_grids
+<br><br>
+
+## generate_crop_boxes
+---
+### ::: ultralytics.vit.sam.amg.generate_crop_boxes
+<br><br>
+
+## uncrop_boxes_xyxy
+---
+### ::: ultralytics.vit.sam.amg.uncrop_boxes_xyxy
+<br><br>
+
+## uncrop_points
+---
+### ::: ultralytics.vit.sam.amg.uncrop_points
+<br><br>
+
+## uncrop_masks
+---
+### ::: ultralytics.vit.sam.amg.uncrop_masks
+<br><br>
+
+## remove_small_regions
+---
+### ::: ultralytics.vit.sam.amg.remove_small_regions
+<br><br>
+
+## coco_encode_rle
+---
+### ::: ultralytics.vit.sam.amg.coco_encode_rle
+<br><br>
+
+## batched_mask_to_box
+---
+### ::: ultralytics.vit.sam.amg.batched_mask_to_box
+<br><br>

docs/reference/vit/sam/autosize.md

@@ -0,0 +1,9 @@
+---
+description: Learn how to use the ResizeLongestSide module in Ultralytics YOLO for automatic image resizing. Resize your images with ease.
+keywords: ResizeLongestSide, Ultralytics YOLO, automatic image resizing, image resizing
+---
+
+## ResizeLongestSide
+---
+### ::: ultralytics.vit.sam.autosize.ResizeLongestSide
+<br><br>

docs/reference/vit/sam/build.md

@@ -0,0 +1,29 @@
+---
+description: Learn how to build SAM and VIT models with Ultralytics YOLO Docs. Enhance your understanding of computer vision models today!.
+keywords: SAM, VIT, computer vision models, build SAM models, build VIT models, Ultralytics YOLO Docs
+---
+
+## build_sam_vit_h
+---
+### ::: ultralytics.vit.sam.build.build_sam_vit_h
+<br><br>
+
+## build_sam_vit_l
+---
+### ::: ultralytics.vit.sam.build.build_sam_vit_l
+<br><br>
+
+## build_sam_vit_b
+---
+### ::: ultralytics.vit.sam.build.build_sam_vit_b
+<br><br>
+
+## _build_sam
+---
+### ::: ultralytics.vit.sam.build._build_sam
+<br><br>
+
+## build_sam
+---
+### ::: ultralytics.vit.sam.build.build_sam
+<br><br>

docs/reference/vit/sam/model.md

@@ -0,0 +1,9 @@
+---
+description: Learn about the Ultralytics VIT SAM model for object detection and how it can help streamline your computer vision workflow. Check out the documentation for implementation details and examples.
+keywords: Ultralytics, VIT, SAM, object detection, computer vision, deep learning, implementation, examples
+---
+
+## SAM
+---
+### ::: ultralytics.vit.sam.model.SAM
+<br><br>

docs/reference/vit/sam/modules/decoders.md

@@ -0,0 +1,9 @@
+## MaskDecoder
+---
+### ::: ultralytics.vit.sam.modules.decoders.MaskDecoder
+<br><br>
+
+## MLP
+---
+### ::: ultralytics.vit.sam.modules.decoders.MLP
+<br><br>

docs/reference/vit/sam/modules/encoders.md

@@ -0,0 +1,54 @@
+---
+description: Learn about Ultralytics ViT encoder, position embeddings, attention, window partition, and more in our comprehensive documentation.
+keywords: Ultralytics YOLO, ViT Encoder, Position Embeddings, Attention, Window Partition, Rel Pos Encoding
+---
+
+## ImageEncoderViT
+---
+### ::: ultralytics.vit.sam.modules.encoders.ImageEncoderViT
+<br><br>
+
+## PromptEncoder
+---
+### ::: ultralytics.vit.sam.modules.encoders.PromptEncoder
+<br><br>
+
+## PositionEmbeddingRandom
+---
+### ::: ultralytics.vit.sam.modules.encoders.PositionEmbeddingRandom
+<br><br>
+
+## Block
+---
+### ::: ultralytics.vit.sam.modules.encoders.Block
+<br><br>
+
+## Attention
+---
+### ::: ultralytics.vit.sam.modules.encoders.Attention
+<br><br>
+
+## PatchEmbed
+---
+### ::: ultralytics.vit.sam.modules.encoders.PatchEmbed
+<br><br>
+
+## window_partition
+---
+### ::: ultralytics.vit.sam.modules.encoders.window_partition
+<br><br>
+
+## window_unpartition
+---
+### ::: ultralytics.vit.sam.modules.encoders.window_unpartition
+<br><br>
+
+## get_rel_pos
+---
+### ::: ultralytics.vit.sam.modules.encoders.get_rel_pos
+<br><br>
+
+## add_decomposed_rel_pos
+---
+### ::: ultralytics.vit.sam.modules.encoders.add_decomposed_rel_pos
+<br><br>

docs/reference/vit/sam/modules/mask_generator.md

@@ -0,0 +1,9 @@
+---
+description: Learn about the SamAutomaticMaskGenerator module in Ultralytics YOLO, an automatic mask generator for image segmentation.
+keywords: SamAutomaticMaskGenerator, Ultralytics YOLO, automatic mask generator, image segmentation
+---
+
+## SamAutomaticMaskGenerator
+---
+### ::: ultralytics.vit.sam.modules.mask_generator.SamAutomaticMaskGenerator
+<br><br>

docs/reference/vit/sam/modules/prompt_predictor.md

@@ -0,0 +1,9 @@
+---
+description: Learn about PromptPredictor - a module in Ultralytics VIT SAM that predicts image captions based on prompts. Get started today!.
+keywords: PromptPredictor, Ultralytics, YOLO, VIT SAM, image captioning, deep learning, computer vision
+---
+
+## PromptPredictor
+---
+### ::: ultralytics.vit.sam.modules.prompt_predictor.PromptPredictor
+<br><br>

docs/reference/vit/sam/modules/sam.md

@@ -0,0 +1,9 @@
+---
+description: Explore the Sam module in Ultralytics VIT, a PyTorch-based vision library, and learn how to improve your image classification and segmentation tasks.
+keywords: Ultralytics VIT, Sam module, PyTorch vision library, image classification, segmentation tasks
+---
+
+## Sam
+---
+### ::: ultralytics.vit.sam.modules.sam.Sam
+<br><br>

docs/reference/vit/sam/modules/transformer.md

@@ -0,0 +1,19 @@
+---
+description: Explore the Attention and TwoWayTransformer modules in Ultralytics YOLO documentation. Learn how to integrate them in your project efficiently.
+keywords: Ultralytics YOLO, Attention module, TwoWayTransformer module, Object Detection, Deep Learning
+---
+
+## TwoWayTransformer
+---
+### ::: ultralytics.vit.sam.modules.transformer.TwoWayTransformer
+<br><br>
+
+## TwoWayAttentionBlock
+---
+### ::: ultralytics.vit.sam.modules.transformer.TwoWayAttentionBlock
+<br><br>
+
+## Attention
+---
+### ::: ultralytics.vit.sam.modules.transformer.Attention
+<br><br>

docs/reference/vit/sam/predict.md

@@ -0,0 +1,9 @@
+---
+description: The VIT SAM Predictor from Ultralytics provides object detection capabilities for YOLO. Learn how to use it and speed up your object detection models.
+keywords: Ultralytics, VIT SAM Predictor, object detection, YOLO
+---
+
+## Predictor
+---
+### ::: ultralytics.vit.sam.predict.Predictor
+<br><br>

docs/reference/vit/utils/loss.md

@@ -0,0 +1,14 @@
+---
+description: DETRLoss is a method for optimizing detection of objects in images. Learn how to use it in RTDETRDetectionLoss at Ultralytics Docs.
+keywords: DETRLoss, RTDETRDetectionLoss, Ultralytics, object detection, image classification, computer vision
+---
+
+## DETRLoss
+---
+### ::: ultralytics.vit.utils.loss.DETRLoss
+<br><br>
+
+## RTDETRDetectionLoss
+---
+### ::: ultralytics.vit.utils.loss.RTDETRDetectionLoss
+<br><br>

docs/reference/vit/utils/ops.md

@@ -0,0 +1,19 @@
+---
+description: Learn about HungarianMatcher and inverse_sigmoid functions in the Ultralytics YOLO Docs. Improve your object detection skills today!.
+keywords: Ultralytics, YOLO, object detection, HungarianMatcher, inverse_sigmoid
+---
+
+## HungarianMatcher
+---
+### ::: ultralytics.vit.utils.ops.HungarianMatcher
+<br><br>
+
+## get_cdn_group
+---
+### ::: ultralytics.vit.utils.ops.get_cdn_group
+<br><br>
+
+## inverse_sigmoid
+---
+### ::: ultralytics.vit.utils.ops.inverse_sigmoid
+<br><br>

mkdocs.yml

@@ -273,6 +273,28 @@ nav:
               - gmc: reference/tracker/utils/gmc.md
               - kalman_filter: reference/tracker/utils/kalman_filter.md
               - matching: reference/tracker/utils/matching.md
+      - vit:
+          - rtdetr:
+              - model: reference/vit/rtdetr/model.md
+              - predict: reference/vit/rtdetr/predict.md
+              - train: reference/vit/rtdetr/train.md
+              - val: reference/vit/rtdetr/val.md
+          - sam:
+              - amg: reference/vit/sam/amg.md
+              - autosize: reference/vit/sam/autosize.md
+              - build: reference/vit/sam/build.md
+              - model: reference/vit/sam/model.md
+              - modules:
+                  - decoders: reference/vit/sam/modules/decoders.md
+                  - encoders: reference/vit/sam/modules/encoders.md
+                  - mask_generator: reference/vit/sam/modules/mask_generator.md
+                  - prompt_predictor: reference/vit/sam/modules/prompt_predictor.md
+                  - sam: reference/vit/sam/modules/sam.md
+                  - transformer: reference/vit/sam/modules/transformer.md
+              - predict: reference/vit/sam/predict.md
+          - utils:
+              - loss: reference/vit/utils/loss.md
+              - ops: reference/vit/utils/ops.md
       - yolo:
           - cfg:
               - __init__: reference/yolo/cfg/__init__.md

ultralytics/vit/rtdetr/model.py

@@ -1,6 +1,6 @@
 # Ultralytics YOLO 🚀, AGPL-3.0 license
 """
-# RT-DETR model interface
+RT-DETR model interface
 """
 
 from pathlib import Path

ultralytics/vit/rtdetr/train.py

@@ -1,3 +1,5 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license
+
 from copy import copy
 
 import torch

ultralytics/vit/sam/__init__.py

@@ -1,3 +1,5 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license
+
 from .build import build_sam  # noqa
 from .model import SAM  # noqa
 from .modules.prompt_predictor import PromptPredictor  # noqa

ultralytics/vit/sam/amg.py

@@ -1,3 +1,5 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license
+
 import math
 from copy import deepcopy
 from itertools import product

ultralytics/vit/sam/autosize.py

@@ -1,3 +1,5 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license
+
 # Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
 

ultralytics/vit/sam/build.py

@@ -1,3 +1,5 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license
+
 # Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
 

ultralytics/vit/sam/model.py

@@ -1,4 +1,7 @@
-# SAM model interface
+# Ultralytics YOLO 🚀, AGPL-3.0 license
+"""
+SAM model interface
+"""
 
 from ultralytics.yolo.cfg import get_cfg
 

ultralytics/vit/sam/modules/__init__.py

@@ -0,0 +1 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license

ultralytics/vit/sam/modules/decoders.py

@@ -1,3 +1,5 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license
+
 from typing import List, Tuple, Type
 
 import torch

ultralytics/vit/sam/modules/encoders.py

@@ -1,3 +1,5 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license
+
 from typing import Any, Optional, Tuple, Type
 
 import numpy as np

ultralytics/vit/sam/modules/mask_generator.py

@@ -1,3 +1,5 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license
+
 # Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
 

ultralytics/vit/sam/modules/prompt_predictor.py

@@ -1,3 +1,5 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license
+
 from typing import Optional, Tuple
 
 import numpy as np

ultralytics/vit/sam/modules/sam.py

@@ -1,3 +1,5 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license
+
 # Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
 

ultralytics/vit/sam/modules/transformer.py

@@ -1,3 +1,5 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license
+
 import math
 from typing import Tuple, Type
 

ultralytics/vit/sam/predict.py

@@ -1,3 +1,5 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license
+
 import numpy as np
 import torch
 

ultralytics/vit/utils/__init__.py

@@ -0,0 +1 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license

ultralytics/vit/utils/loss.py

@@ -1,3 +1,5 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license
+
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
@@ -18,11 +20,12 @@ class DETRLoss(nn.Module):
                  use_uni_match=False,
                  uni_match_ind=0):
         """
+        DETR loss function.
+
         Args:
             nc (int): The number of classes.
             loss_gain (dict): The coefficient of loss.
             aux_loss (bool): If 'aux_loss = True', loss at each decoder layer are to be used.
-            use_focal_loss (bool): Use focal loss or not.
             use_vfl (bool): Use VarifocalLoss or not.
             use_uni_match (bool): Whether to use a fixed layer to assign labels for auxiliary branch.
             uni_match_ind (int): The fixed indices of a layer.

ultralytics/vit/utils/ops.py

@@ -1,4 +1,4 @@
-# TODO: license
+# Ultralytics YOLO 🚀, AGPL-3.0 license
 
 import torch
 import torch.nn as nn
@@ -10,12 +10,31 @@ from ultralytics.yolo.utils.ops import xywh2xyxy, xyxy2xywh
 
 
 class HungarianMatcher(nn.Module):
+    """
+    A module implementing the HungarianMatcher, which is a differentiable module to solve the assignment problem in
+    an end-to-end fashion.
+
+    HungarianMatcher performs optimal assignment over predicted and ground truth bounding boxes using a cost function
+    that considers classification scores, bounding box coordinates, and optionally, mask predictions.
+
+    Attributes:
+        cost_gain (dict): Dictionary of cost coefficients for different components: 'class', 'bbox', 'giou', 'mask', and 'dice'.
+        use_fl (bool): Indicates whether to use Focal Loss for the classification cost calculation.
+        with_mask (bool): Indicates whether the model makes mask predictions.
+        num_sample_points (int): The number of sample points used in mask cost calculation.
+        alpha (float): The alpha factor in Focal Loss calculation.
+        gamma (float): The gamma factor in Focal Loss calculation.
+
+    Methods:
+        forward(pred_bboxes, pred_scores, gt_bboxes, gt_cls, gt_groups, masks=None, gt_mask=None): Computes the assignment
+        between predictions and ground truths for a batch.
+        _cost_mask(bs, num_gts, masks=None, gt_mask=None): Computes the mask cost and dice cost if masks are predicted.
+    """
+
+    class HungarianMatcher(nn.Module):
+        ...
 
     def __init__(self, cost_gain=None, use_fl=True, with_mask=False, num_sample_points=12544, alpha=0.25, gamma=2.0):
-        """
-        Args:
-            matcher_coeff (dict): The coefficient of hungarian matcher cost.
-        """
         super().__init__()
         if cost_gain is None:
             cost_gain = {'class': 1, 'bbox': 5, 'giou': 2, 'mask': 1, 'dice': 1}
@@ -28,22 +47,30 @@ class HungarianMatcher(nn.Module):
 
     def forward(self, pred_bboxes, pred_scores, gt_bboxes, gt_cls, gt_groups, masks=None, gt_mask=None):
         """
+        Forward pass for HungarianMatcher. This function computes costs based on prediction and ground truth
+        (classification cost, L1 cost between boxes and GIoU cost between boxes) and finds the optimal matching
+        between predictions and ground truth based on these costs.
+
         Args:
-            pred_bboxes (Tensor): [b, query, 4]
+            pred_bboxes (Tensor): Predicted bounding boxes with shape [batch_size, num_queries, 4].
-            pred_scores (Tensor): [b, query, num_classes]
+            pred_scores (Tensor): Predicted scores with shape [batch_size, num_queries, num_classes].
-            gt_cls (torch.Tensor) with shape [num_gts, ]
+            gt_cls (torch.Tensor): Ground truth classes with shape [num_gts, ].
-            gt_bboxes (torch.Tensor): [num_gts, 4]
+            gt_bboxes (torch.Tensor): Ground truth bounding boxes with shape [num_gts, 4].
-            gt_groups (List(int)): a list of batch size length includes the number of gts of each image.
+            gt_groups (List[int]): List of length equal to batch size, containing the number of ground truths for
-            masks (Tensor|None): [b, query, h, w]
+                each image.
-            gt_mask (List(Tensor)): list[[n, H, W]]
+            masks (Tensor, optional): Predicted masks with shape [batch_size, num_queries, height, width].
+                Defaults to None.
+            gt_mask (List[Tensor], optional): List of ground truth masks, each with shape [num_masks, Height, Width].
+                Defaults to None.
 
         Returns:
-            A list of size batch_size, containing tuples of (index_i, index_j) where:
+            (List[Tuple[Tensor, Tensor]]): A list of size batch_size, each element is a tuple (index_i, index_j), where:
-                - index_i is the indices of the selected predictions (in order)
+                - index_i is the tensor of indices of the selected predictions (in order)
-                - index_j is the indices of the corresponding selected targets (in order)
+                - index_j is the tensor of indices of the corresponding selected ground truth targets (in order)
-            For each batch element, it holds:
+                For each batch element, it holds:
-                len(index_i) = len(index_j) = min(num_queries, num_target_boxes)
+                    len(index_i) = len(index_j) = min(num_queries, num_target_boxes)
         """
+
         bs, nq, nc = pred_scores.shape
 
         if sum(gt_groups) == 0:
@@ -124,24 +151,29 @@ def get_cdn_group(batch,
                   cls_noise_ratio=0.5,
                   box_noise_scale=1.0,
                   training=False):
-    """Get contrastive denoising training group
+    """
+    Get contrastive denoising training group. This function creates a contrastive denoising training group with
+    positive and negative samples from the ground truths (gt). It applies noise to the class labels and bounding
+    box coordinates, and returns the modified labels, bounding boxes, attention mask and meta information.
 
     Args:
-        batch (dict): A dict includes:
+        batch (dict): A dict that includes 'gt_cls' (torch.Tensor with shape [num_gts, ]), 'gt_bboxes'
-            gt_cls (torch.Tensor) with shape [num_gts, ],
+            (torch.Tensor with shape [num_gts, 4]), 'gt_groups' (List(int)) which is a list of batch size length
-            gt_bboxes (torch.Tensor): [num_gts, 4],
+            indicating the number of gts of each image.
-            gt_groups (List(int)): a list of batch size length includes the number of gts of each image.
         num_classes (int): Number of classes.
         num_queries (int): Number of queries.
-        class_embed (torch.Tensor): Embedding weights to map cls to embedding space.
+        class_embed (torch.Tensor): Embedding weights to map class labels to embedding space.
-        num_dn (int): Number of denoising.
+        num_dn (int, optional): Number of denoising. Defaults to 100.
-        cls_noise_ratio (float): Noise ratio for class.
+        cls_noise_ratio (float, optional): Noise ratio for class labels. Defaults to 0.5.
-        box_noise_scale (float): Noise scale for bbox.
+        box_noise_scale (float, optional): Noise scale for bounding box coordinates. Defaults to 1.0.
-        training (bool): If it's training or not.
+        training (bool, optional): If it's in training mode. Defaults to False.
 
     Returns:
-
+        (Tuple[Optional[Tensor], Optional[Tensor], Optional[Tensor], Optional[Dict]]): The modified class embeddings,
+            bounding boxes, attention mask and meta information for denoising. If not in training mode or 'num_dn'
+            is less than or equal to 0, the function returns None for all elements in the tuple.
     """
+
     if (not training) or num_dn <= 0:
         return None, None, None, None
     gt_groups = batch['gt_groups']