AI in agriculture is only as good as the data behind it, but messy datasets, poor annotations, and hidden biases slow progress. Join Paula for a dynamic session on semantic segmentation, where she will show how FiftyOne can transform dataset curation, annotation analysis, and model evaluation for agricultural AI. Using real-world coffee datasets from Colombia, we’ll dive into the segmentation of coffee fruits at different maturation stages, leveraging FiftyOne’s powerful tools, from uniqueness detection to similarity search and embedding visualizations. Whether interested in farm robotics, remote sensing, or plant phenotyping, this talk will give you actionable techniques to refine your datasets and supercharge your AI workflows.
#computervision #ai #artificialintelligence #machinevision #machinelearning #datascience #opensource
Contents of this video --
00:00 - Introduction
00:52 - Background and Mission in AI for Agriculture
01:48 - Importance of Coffee in Colombia
02:54 - Empowering Farmers with Mobile Tools
03:43 - Data Collection and Device Design
04:47 - Challenges in Coffee Yield Estimation
05:52 - Project Timeline and Annotation Process
07:05 - Computer Vision and 3D Analysis Techniques
08:01 - Revisiting Past Models with New Tools
08:34 - Introducing the Coffee Dataset with FiftyOne
09:22 - Using the FiftyOne App for Dataset Inspection
11:04 - Understanding Embeddings and Outliers
12:01 - Improving Dataset Quality with FiftyOne
13:33 - Data-Centric AI vs. Model-Centric AI
14:09 - Loading and Exploring Data with FiftyOne SDK
15:04 - Interactive Geolocation Plot for Agriculture
17:02 - Visualizing Regional Coffee Data
17:51 - Dataset Support and Use Cases in FiftyOne
18:31 - Using SAM-2 with FiftyOne’s Model Zoo
20:26 - Pre-Annotation Strategy with Unique Samples
22:01 - Annotation Integration and CAT Tool
23:36 - Training Model and Achieving 90% Accuracy
24:54 - Embracing a Data-Centric AI Approach
26:02 - Visual AI Benefits with FiftyOne
27:36 - Final Call to Action and Resources
28:13 - Audience Q&A: Unique vs. Large Datasets
--
#computervision #ai #artificialintelligence #machinevision #machinelearning #datascience #opensource