links

Author: nibaccam

articles/machine-learning/how-to-auto-train-image-models.md

@@ -186,7 +186,7 @@ automl_image_config = AutoMLImageConfig(compute_target=compute_target)
 
 With support for computer vision tasks, you can control the model algorithm and sweep hyperparameters. These model algorithms and hyperparameters are passed in as the parameter space for the sweep.
 
-The model algorithm is required and is passed in via `model_name` parameter. You can either specify a single `model_name` or choose between multiple. In addition to controlling the model algorithm, you can also tune hyperparameters used for model training. While many of the hyperparameters exposed are model-agnostic, there are instances where hyperparameters are task-specific or model-specific. [Learn more about the available hyperparameters for these instances](reference-automl-images-hyperparameter.md). 
+The model algorithm is required and is passed in via `model_name` parameter. You can either specify a single `model_name` or choose between multiple. In addition to controlling the model algorithm, you can also tune hyperparameters used for model training. While many of the hyperparameters exposed are model-agnostic, there are instances where hyperparameters are task-specific or model-specific. [Learn more about the available hyperparameters for these instances](reference-automl-images-hyperparameters.md). 
 
 ### Supported model algorithms
 
@@ -195,7 +195,7 @@ The following table summarizes the supported models for each computer vision tas
 Task |  Model algorithms | String literal syntax<br> ***`default_model`\**** denoted with \*
 ---|----------|----------
 Image classification<br> (multi-class and multi-label)| **MobileNet**: Light-weighted models for mobile applications <br> **ResNet**: Residual networks<br> **ResNeSt**: Split attention networks<br> **SE-ResNeXt50**: Squeeze-and-Excitation networks<br> **ViT**: Vision transformer networks| `mobilenetv2`   <br>`resnet18` <br>`resnet34` <br> `resnet50`  <br> `resnet101` <br> `resnet152`    <br> `resnest50` <br> `resnest101`  <br> `seresnext`  <br> `vits16r224` (small) <br> ***`vitb16r224`\**** (base) <br>`vitl16r224` (large)|
-Object detection | **YOLOv5**: One stage object detection model   <br>  **Faster RCNN ResNet FPN**: Two stage object detection models  <br> **RetinaNet ResNet FPN**: address class imbalance with Focal Loss <br> <br>*Note: Refer to [`model_size` hyperparameter](#model-specific-hyperparameters) for YOLOv5 model sizes.*| ***`yolov5`\**** <br> `fasterrcnn_resnet18_fpn` <br> `fasterrcnn_resnet34_fpn` <br> `fasterrcnn_resnet50_fpn` <br> `fasterrcnn_resnet101_fpn` <br> `fasterrcnn_resnet152_fpn` <br> `retinanet_resnet50_fpn` 
+Object detection | **YOLOv5**: One stage object detection model   <br>  **Faster RCNN ResNet FPN**: Two stage object detection models  <br> **RetinaNet ResNet FPN**: address class imbalance with Focal Loss <br> <br>*Note: Refer to [`model_size` hyperparameter](reference-automl-images-hyperparameters.md#model-specific-hyperparameters) for YOLOv5 model sizes.*| ***`yolov5`\**** <br> `fasterrcnn_resnet18_fpn` <br> `fasterrcnn_resnet34_fpn` <br> `fasterrcnn_resnet50_fpn` <br> `fasterrcnn_resnet101_fpn` <br> `fasterrcnn_resnet152_fpn` <br> `retinanet_resnet50_fpn` 
 Instance segmentation | **MaskRCNN ResNet FPN**| `maskrcnn_resnet18_fpn` <br> `maskrcnn_resnet34_fpn` <br> ***`maskrcnn_resnet50_fpn`\****  <br> `maskrcnn_resnet101_fpn` <br> `maskrcnn_resnet152_fpn` <br>`maskrcnn_resnet50_fpn`
 
 ### Data augmentation 
@@ -491,7 +491,7 @@ Each of the tasks (and some models) have a set of parameters in the `model_setti
 |Object detection, instance segmentation| `min_size`<br>`max_size`<br>`box_score_thresh`<br>`box_nms_thresh`<br>`box_detections_per_img` | 600<br>1333<br>0.3<br>0.5<br>100 |
 |Object detection using `yolov5`| `img_size`<br>`model_size`<br>`box_score_thresh`<br>`box_iou_thresh` | 640<br>medium<br>0.1<br>0.5 |
 
-For a detailed description on these parameters, please refer to the above section on [task specific hyperparameters](#task-specific-hyperparameters).
+For a detailed description on task specific hyperparameters, please refer to [Hyperparameters for computer vision tasks in automated machine learning](reference-automl-images-hyperparameters.md).
 
 If you want to use tiling, and want to control tiling behavior, the following parameters are available: `tile_grid_size`, `tile_overlap_ratio` and `tile_predictions_nms_thresh`. For more details on these parameters please check [Train a small object detection model using AutoML](how-to-use-automl-small-object-detect.md).
 

articles/machine-learning/how-to-understand-automated-ml.md

@@ -262,10 +262,10 @@ Please refer to the metrics definitions from the [classification metrics](#class
 
 ![Classification report for image classification](./media/how-to-understand-automated-ml/image-classification-report.png)
 
-### Object detection and Instance segmentation metrics
+### Object detection and instance segmentation metrics
 
 Every prediction from an image object detection or instance segmentation  model is associated with a confidence score.
-The predictions with confidence score greater than score threshold are output as predictions and used in the metric calculation, the default value of which is model specific and can be referred from the [hyperparameter tuning](how-to-auto-train-image-models.md#model-specific-hyperparameters) page(`box_score_threshold` hyperparameter).
+The predictions with confidence score greater than score threshold are output as predictions and used in the metric calculation, the default value of which is model specific and can be referred from the [hyperparameter tuning](reference-automl-images-hyperparameters.md#model-specific-hyperparameters) page(`box_score_threshold` hyperparameter).
 
 The metric computation of an image object detection and instance segmentation model is based on an overlap measurement defined by a metric called **IoU** ([Intersection over Union](https://en.wikipedia.org/wiki/Jaccard_index)) which is computed by dividing the area of overlap between the ground-truth and the predictions by the area of union of the ground-truth and the predictions. The IoU computed from every prediction is compared with an **overlap threshold** called an IoU threshold which determines how much a prediction should overlap with a user-annotated ground-truth in order to be considered as a positive prediction. If the IoU computed from the prediction is less than the overlap threshold the prediction would not be considered as a positive prediction for the associated class.
 
@@ -281,7 +281,7 @@ The primary metric for the evaluation of image object detection and instance seg
 [COCO evaluation method](https://cocodataset.org/#detection-eval) uses a 101-point interpolated method for AP calculation along with averaging over ten IoU thresholds. AP@[.5:.95] corresponds to the average AP for IoU from 0.5 to 0.95 with a step size of 0.05. Automated ML logs all the twelve metrics defined by the COCO method including the AP and AR(average recall) at various scales in the application logs while the metrics user interface shows only the mAP  at an IoU threshold of 0.5. 
 
 > [!TIP]
-> The image object detection model evaluation can use coco metrics if the `validation_metric_type` hyperparameter is set to be 'coco' as explained in the [hyperparameter tuning](how-to-auto-train-image-models.md#task-specific-hyperparameters) section.
+> The image object detection model evaluation can use coco metrics if the `validation_metric_type` hyperparameter is set to be 'coco' as explained in the [hyperparameter tuning](reference-automl-images-hyperparameters.md#object-detection-and-instance-segmentation-task-specific-hyperparameters) section.
 
 #### Epoch-level metrics for object detection and instance segmentation
 The mAP, precision and recall values are logged at an epoch-level for image object detection/instance segmentation models. The mAP, precision and recall metrics are also logged at a class level with the name 'per_label_metrics'. The 'per_label_metrics' should be viewed as a table.