Dávid Komorowicz
@dawars.me
PhD candidate @Jena_DH & @TU_Muenchen working on 3D Reconstruction from Historic Imagery. @TU_Muenchen graduate.
📍 Munich
📍 Munich
I completely failed at my coding interview so I'm starting over from the basics. 😅
February 21, 2025 at 10:06 AM
I completely failed at my coding interview so I'm starting over from the basics. 😅
Using multi modal foundation models
November 19, 2024 at 6:57 PM
Using multi modal foundation models
Latest papers:
- Gravity aligned rotation averaging with circular regression ECCV24
- Breaking the Frame: Visual Place Recognition by Overlap Prediction WACV25
- Gravity aligned rotation averaging with circular regression ECCV24
- Breaking the Frame: Visual Place Recognition by Overlap Prediction WACV25
November 19, 2024 at 5:46 PM
Latest papers:
- Gravity aligned rotation averaging with circular regression ECCV24
- Breaking the Frame: Visual Place Recognition by Overlap Prediction WACV25
- Gravity aligned rotation averaging with circular regression ECCV24
- Breaking the Frame: Visual Place Recognition by Overlap Prediction WACV25
Experiments and datasets:
- more cameras correctly registered
- way faster for large datasets
Downstream effect of pose quality in NeRFs
- more cameras correctly registered
- way faster for large datasets
Downstream effect of pose quality in NeRFs
November 19, 2024 at 5:46 PM
Experiments and datasets:
- more cameras correctly registered
- way faster for large datasets
Downstream effect of pose quality in NeRFs
- more cameras correctly registered
- way faster for large datasets
Downstream effect of pose quality in NeRFs
In the global approach, cameras are optimized at the same time.
1. Rotation averaging: relative -> absolute optimization
2. ❌ Translation averaging (hard) unknown translation, few constraints
2. ✅ Global positioning: camera-to-point constraints
1. Rotation averaging: relative -> absolute optimization
2. ❌ Translation averaging (hard) unknown translation, few constraints
2. ✅ Global positioning: camera-to-point constraints
November 19, 2024 at 5:46 PM
In the global approach, cameras are optimized at the same time.
1. Rotation averaging: relative -> absolute optimization
2. ❌ Translation averaging (hard) unknown translation, few constraints
2. ✅ Global positioning: camera-to-point constraints
1. Rotation averaging: relative -> absolute optimization
2. ❌ Translation averaging (hard) unknown translation, few constraints
2. ✅ Global positioning: camera-to-point constraints
Initializing pose graph is hard because translation is unknown.
Same for glomap and colmap:
1. Finding overlapping images using retrieval
2. Find matching points across image pairs
3. Estimate relative poses robustly
Same for glomap and colmap:
1. Finding overlapping images using retrieval
2. Find matching points across image pairs
3. Estimate relative poses robustly
November 19, 2024 at 5:46 PM
Initializing pose graph is hard because translation is unknown.
Same for glomap and colmap:
1. Finding overlapping images using retrieval
2. Find matching points across image pairs
3. Estimate relative poses robustly
Same for glomap and colmap:
1. Finding overlapping images using retrieval
2. Find matching points across image pairs
3. Estimate relative poses robustly
Starting ZurichCV with Global Structure from Motion by Dániel Baráth
November 19, 2024 at 5:10 PM
Starting ZurichCV with Global Structure from Motion by Dániel Baráth