Object Detection with KerasCV

on YouTube ↗22m

" summary

KerasCV provides a full object detection pipeline that trains a RetinaNet on custom datasets, with native support for ragged bounding boxes, TPU‑compatible training loops, and COCO metrics implemented in the computation graph. The demo shows composable augmentations like mix up and jittered resize, a worked notebook that runs with model.fit, and predictions improving after a few epochs. The claim is concrete: ready‑to‑use RetinaNet training, modular components, and working evaluation inside Keras.

AI-written summary · about

" chapters (15)

1. 0:00Introduction
2. 0:50Motivation for KerasCV object detection
3. 3:39Object detection challenges
4. 5:27API design and bounding boxes
5. 7:34Augmentation pipelines with KerasCV
6. 9:19Building RetinaNet models
7. 10:40Evaluating metrics with COCO
8. 11:43Simple inference and deployment
9. 12:27Live demo setup
10. 13:09Loading and preprocessing data
11. 14:13Augmentation layers in action
12. 17:20Training the object detection model
13. 18:46Model predictions and results
14. 19:27Configuring non-max suppression
15. 20:29Summary and future

" related talks

• 01h 11mObject detection using OpenCVAmir Siddiqui · DSC Europe 23
• 28mObject Detection using Transformers and CNNsEduardo Dixo · Conf42 Machine Learning 2021
• 01h 29mComputer Vision: An (Un?)Expected Journey, with Keras and TensorflowRodolfo Bonnin · PyData London 2018
• 01h 07mComputer Vision with PyTorchDaniel J. Brooks · PyData Los Angeles 2019
• 36mObject detection to Instance segmentation: Learn to apply several algorithms...Sujatha Subramanian · PyData Berlin 2018
• 12mCascade models in computer vision to boost accuracy and performanceArgo Saakyan · Conf42 Machine Learning 2023
• 01h 15mTrain Object Detection with Small DatasetsVincenzo Crescimanna · PyData London 2022
• 41mPractical Aspects of Applying Modern Computer Vision TechnologiesDaria Lashkevich · Big Data Conference Europe 2022