HOW TO GUIDE

How to filter DeepSparse detections

Before you can train a computer vision model, you need labeled data on which to train your model. The more accurate the labels, or annotations, are, the higher the performance the model will achieve.
Get started for free

Overview

Before a model goes to production, you need to build logic around the model. This often involves filtering detections from the model. For instance, you may want to retrieve detections from a model that are above a specified confidence level, or detections in a particular region of an image or video.

Using supervision, you can filter detections from computer vision models in a few lines of code.

supervision is an open-source Python package with a range of utilities for working with computer vision models. For this tutorial, we will focus specifically on filtering detections with supervision, but you can also use supervision for model evaluation, visualizing detections, and more.

Below, we show how to filter

DeepSparse

detections with supervision. We will:

1. Install supervision
2. Load

DeepSparse

model detections
3. Show how to filter by confidence, area, and more

Without further ado, let's get started!

DeepSparse and Image Annotation Resources

Explore these resources to enhance your understanding of DeepSparse and image annotation techniques.
Step 1

Install supervision

First, install the supervision pip package: 

pip install supervision


Once you have installed supervision, you are ready to load your data and start writing logic to filter detections.

Step 2

Load Data

supervision has a standard class called supervision.Detections() that provides a range of functions for working with bounding box and segmentation detections.

You can load predictions into a Detections() object in a single line of code using the supervision "from" loaders.

For

DeepSparse

predictions, we will use the

data loader:
‍


from deepsparse import Pipeline
import supervision as sv

yolo_pipeline = Pipeline.create(
    task="yolo",
    model_path="zoo:cv/detection/yolov5-l/pytorch/ultralytics/coco/pruned80_quant-none"
)
pipeline_outputs = yolo_pipeline(SOURCE_IMAGE_PATH, iou_thres=0.6, conf_thres=0.001)
detections = sv.Detections.from_deepsparse(pipeline_outputs)
Step 3

With supervision, you can filter detections by:

  1. A specific class
  2. A set of classes
  3. Confidence
  4. Bounding box area
  5. Box dimensions
  6. A specific zone

You can combine these filters to build the logic you need.

For this guide, we will focus on filtering by classes and confidence. To learn about the other filters available in supervision, check out the Detections() quickstart guide.

Filter by Class

To filter

DeepSparse

detections by class, use the following code:


detections = detections[detections.class_id == 0]


Replace the number 0 with the ID of the class whose predictions you want to retriev.e

Filter by Set of Classes

To retrieve

DeepSparse

detections by class, use the following code:


selected_classes = [0, 2, 3]
detections = detections[np.isin(detections.class_id, selected_classes)]

Where selected_classes contains the list of class IDs you want to filter.

Filter by Confidence

To retrieve

DeepSparse

detections over a specified confidence level, use the following code:


detections = detections[detections.confidence > 0.5]
Step 4

Next Steps

supervision provides an extensive range of functionalities for working with computer vision models. With supervision, you can:

1. Process and filter detections and segmentation masks from a range of popular models (YOLOv5, Ultralytics YOLOv8, MMDetection, and more).
2. Process and filter classifications
3. Compute confusion matrices
4. Plot bounding boxes and segmentation masks

And more! To learn about the full range of functionality in supervision, check out the supervision documentation.

Ready to get started?

Join over 1 million developers building with Roboflow

Try it for FreeTalk to Sales