Object Segmentation in Satellite Imagery (Kaggle DSTL)
summary
Presents a solution to the Kaggle Dstl Satellite Imagery Feature Detection challenge: segmenting objects in satellite images. Describes the team approach (with Roman Solovyov), the modeling and training pipeline used to achieve second place in the competition.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 3:20Data problems and metrics
- 5:06Baseline solution
- 7:13U-Net model and improvements
- 9:47Final solution pipeline
- 13:50Error analysis and testing
- 15:59Roman's team solution
- 17:46Other participants' solutions
- 20:34Weights scandal and leaderboard
- 21:16Results and demo
related talks
36mClassification of Satellite Imagery
33mWhat and where? - Machine learning for geospatial image analysis
36mDeep Learning - Segmentation, Identification and Classification
46mDeep Learning for Satellite Image Processing, Lessons Learned
31mFrom Radar Echoes To Real-Time Recognition: Deep Learning on Another Planet -An Open Source Workflow
24mExploring and Analyzing Open Satellite Imagery data
18mBuilding Damage Detection from Satellite Imagery
39mImage Recognition DL WORKSHOP