Plus BIProject Opportunities from TU Munich
Saved projects…Stored in a cookie on this browser. Clearing cookies removes bookmarks.
Back to projects

idp-289 · Organization not specified

Video-Driven World Modeling for Granular Powder Flow Dynamics

Automatically extracted from the linked project document. Check the original PDF before relying on a detail.

Document language: English · Pages: 2 · Extracted: 31 Oct 2026, 03:05

Offer 1

Video-Driven World Modeling for Granular Powder Flow Dynamics

English summary

The project adapts pre-trained 3D foundation models to rotating-drum granular-flow video, using video preprocessing, model specialization strategies and a joint spatiotemporal learning head. Students evaluate 3D reconstruction and material-property inference against experimental baselines and conduct ablation studies. The work includes documenting the adaptation pipeline and releasing reproducible inference and training code; 100% remote work is possible.

Overview

Project details
Video-Driven World Modeling for Granular Powder Flow Dynamics
Source excerpt · Page 1
  • “# Video-Driven World Modeling for Granular Powder Flow Dynamics” Page 1
Organization
Not stated in the PDF
Project formats
  • IDP
    Source excerpt · Page 1
    • “Interdisciplinary project (IDP)” Page 1
Degree levels
Not stated in the PDF
Project goal
Specialize pre-trained 3D foundation models for granular powder flow, jointly recovering temporally consistent 3D powder-bed geometry and inferring physical material properties from monocular video.
Source excerpt · Page 1
  • “Given monocular video of evolving powder surfaces, our model will simultaneously solve two coupled tasks: (1) recover temporally consistent 3D geometry of the powder bed; (2) infer implicit physical characteristics governing the observed flow dynamics.” Page 1

Topics and work

Subjects
  • granular powder flow dynamics
    Source excerpt · Page 1
    • “Video sequences of granular powder flow carry coupled signals of geometry, motion and intrinsic material physics.” Page 1
  • 3D computer vision and reconstruction
    Source excerpt · Page 1
    • “The core idea is to leverage the geometric and visual priors encapsulated in off-the-shelf 3D foundation models” Page 1
  • granular material physics
    Source excerpt · Page 1
    • “material properties such as cohesion, friction and flow rheology” Page 1
  • 3D foundation model adaptation
    Source excerpt · Page 1
    • “This project focuses on specializing pre-trained 3D foundation models for this physical granular system.” Page 1
Application areas
  • rotating-drum granular flow experiments
    Source excerpt · Page 1
    • “Curate and preprocess monocular video sequences from rotating-drum granular flow experiments:” Page 1
Methods and tools
  • monocular video sequences
    Source excerpt · Page 1
    • “Given monocular video of evolving powder surfaces” Page 1
  • pre-trained 3D foundation models
    Source excerpt · Page 1
    • “specializing pre-trained 3D foundation models for this physical granular system” Page 1
  • frame sampling
    Source excerpt · Page 1
    • “frame sampling, ROI cropping, photometric normalization, camera calibration and pseudo-multi-view construction” Page 1
  • ROI cropping
    Source excerpt · Page 1
  • photometric normalization
    Source excerpt · Page 1
    • “photometric normalization” Page 1
  • camera calibration
    Source excerpt · Page 1
    • “camera calibration” Page 1
  • pseudo-multi-view construction
    Source excerpt · Page 1
    • “pseudo-multi-view construction for foundation model adaptation” Page 1
  • lightweight finetuning
    Source excerpt · Page 2
    • “design lightweight finetuning, prompt tuning or LoRA-style specialization strategies” Page 2
  • prompt tuning
    Source excerpt · Page 2
  • LoRA-style specialization
    Source excerpt · Page 2
    • “LoRA-style specialization strategies” Page 2
  • joint spatiotemporal learning head
    Source excerpt · Page 2
    • “Build a joint spatiotemporal learning head on top of the adapted 3D foundation backbone” Page 2
  • experimental baselines
    Source excerpt · Page 2
    • “Evaluate reconstruction quality (geometry fidelity, temporal consistency) and property inference accuracy against experimental baselines” Page 2
  • ablation studies
    Source excerpt · Page 2
    • “conduct ablation studies on foundation model adaptation strategies” Page 2
  • inference and training code
    Source excerpt · Page 2
    • “release reproducible inference and training code” Page 2
Activities
  • Curate and preprocess monocular video sequences from rotating-drum granular flow experiments
    Source excerpt · Page 1
    • “Curate and preprocess monocular video sequences from rotating-drum granular flow experiments:” Page 1
  • Adapt existing pre-trained 3D foundation models to the granular flow domain
    Source excerpt · Page 2
    • “Adapt existing pre-trained 3D foundation models to the granular flow domain:” Page 2
  • Build a joint spatiotemporal learning head
    Source excerpt · Page 2
    • “Build a joint spatiotemporal learning head on top of the adapted 3D foundation backbone” Page 2
  • Evaluate reconstruction quality and property inference accuracy against experimental baselines
    Source excerpt · Page 2
    • “Evaluate reconstruction quality (geometry fidelity, temporal consistency) and property inference accuracy against experimental baselines” Page 2
  • Conduct ablation studies on foundation model adaptation strategies
    Source excerpt · Page 2
    • “conduct ablation studies on foundation model adaptation strategies” Page 2
  • Summarize results
    Source excerpt · Page 2
    • “Summarize results, document the adaptation pipeline, and release reproducible inference and training code for the 3D-aware granular world model.” Page 2
  • Document the adaptation pipeline
    Source excerpt · Page 2
    • “document the adaptation pipeline” Page 2
  • Release reproducible inference and training code
    Source excerpt · Page 2
    • “release reproducible inference and training code” Page 2
Kinds of work
  • Data analysis and ML
    Source excerpt · Page 2
    • “Build a joint spatiotemporal learning head on top of the adapted 3D foundation backbone” Page 2
  • Software development
    Source excerpt · Page 2
    • “release reproducible inference and training code” Page 2
Expected outputs
  • adaptation pipeline documentation
    Source excerpt · Page 2
    • “document the adaptation pipeline” Page 2
  • reproducible inference and training code
    Source excerpt · Page 2
    • “release reproducible inference and training code” Page 2

Requirements

Required skills
Not stated in the PDF
Recommended skills
  • Strong proficiency in Python
    Source excerpt · Page 2
    • “Strong proficiency in Python;” Page 2
  • Solid computer vision background, with hands-on experience in video understanding, multi-view geometry and 3D reconstruction
    Source excerpt · Page 2
    • “Solid computer vision background, with hands-on experience in video understanding, multi-view geometry and 3D reconstruction;” Page 2
  • Proficient in PyTorch for deep learning research and model development
    Source excerpt · Page 2
    • “Proficient in PyTorch for deep learning research and model development;” Page 2
  • Practical experience with large-scale deep learning training: distributed training, dataset pipeline design, checkpoint management and hyperparameter tuning
    Source excerpt · Page 2
    • “Practical experience with large-scale deep learning training: distributed training, dataset pipeline design, checkpoint management and hyperparameter tuning;” Page 2
  • Familiarity with 3D / vision foundation models, including adaptation paradigms like LoRA, prompt tuning and domain alignment
    Source excerpt · Page 2
    • “Familiarity with 3D / vision foundation models (e.g. 3DGS, NeRF foundation models, point-cloud foundation models, vision-LiDAR foundation models), including adaptation paradigms like LoRA, prompt tuning and domain alignment;” Page 2
  • Curiosity for interdisciplinary research bridging 3D computer vision, foundation model adaptation and granular material physics
    Source excerpt · Page 2
    • “Curiosity for interdisciplinary research bridging 3D computer vision, foundation model adaptation and granular material physics.” Page 2
Eligible study fields
Not stated in the PDF
Programming in the project
Central
Source excerpt · Page 2
  • “release reproducible inference and training code for the 3D-aware granular world model.” Page 2
Programming prerequisite
Recommended
Source excerpt · Page 2, Page 2
  • “Strong proficiency in Python;” Page 2
  • “Proficient in PyTorch for deep learning research and model development;” Page 2

Practical details

Team size
Not stated in the PDF
Duration
Not stated in the PDF
Start
Not stated in the PDF
Application deadline
Not stated in the PDF
Work location
100% remote work possible
Source excerpt · Page 1
  • “[100% remote work possible]” Page 1
Location mode
Remote
Source excerpt · Page 1
  • “[100% remote work possible]” Page 1
Working language
Not stated in the PDF

Learning and support

Learning opportunities
  • End-to-end research experience covering video preprocessing, foundation model adaptation, temporal-consistent 3D reconstruction, and visual-to-physical property inference
    Source excerpt · Page 2
    • “You will gain end-to-end research experience covering video preprocessing, foundation model adaptation, temporal-consistent 3D reconstruction, and visual-to-physical property inference.” Page 2
Support offered
Not stated in the PDF

Application and contacts

Contacts
  • Yuan Tan · Yuan.tan@tum.de
    Source excerpt · Page 2
    • “Yuan Tan Gregor-Mendel-Straße 4, 85354 Freising Tel. +49 8161 71-5172 Yuan.tan@tum.de” Page 2
How to apply
Not stated in the PDF
Further information
Not stated in the PDF