Zero-Shot Compositional 3D Scene Reconstruction

Goal: Reconstruct arbitrary scenes into compositional 3D representations from single images or videos.

Problem: Prior work generalizes across distribution shifts, yet is inherently restricted to the asset vocabulary and  concepts introduced during finetuning.

Solution: Instead of finetuning language models directly for inverse graphics, leverage pretrained VLMs zero-shot as high-level planners that orchestrate specialist tool.

Main activities :

• Design novel algorithms for reconstructing scences into compositional 3D representations 
• Implement the approach and experiment with different variations
• Evaluate on state-of-the-art datasets and compare these results 
• Test and change until validation
• Write a report describing progress 

Technical skills and level required : Experience in coding and research 

Languages : English

Relational skills : Collaboration and team work skills

Other valued appreciated :

• Subsidized meals
• Partial reimbursement of public transport costs
• Leave: 7 weeks of annual leave + 10 extra days off due to RTT (statutory reduction in working hours) + possibility of exceptional leave (sick children, moving home, etc.)
• Possibility of teleworking and flexible organization of working hours
• Professional equipment available (videoconferencing, loan of computer equipment, etc.)
• Social, cultural and sports events and activities
• Access to vocational training
• Social security coverage