Computer Vision Research Internship: Image to Sequence Modeling (e.g. Transformers)
About this role
Duration: Minimum 6 months; ideally 9-12 months, depending on the candidate's experience
Scandit gives people superpowers. Whether enabling delivery drivers to make quicker deliveries, matching a patient with their medication, or allowing retailers to make store operations more efficient, our technology automates workflows and provides actionable insights to help businesses in a variety of industries. Join us, as we continue to expand, grow and innovate, and help take Scandit to the next level.
About the Internship
We are offering a research-focused internship aimed at advancing machine learning methods for complex visual understanding tasks. The project centers on deep learning architectures for image-to-sequence modelling, such as Transformers, attention mechanisms, and modern sequence and representation-learning frameworks, to address challenging and highly structured computer vision problems. This project contributes to long-term research efforts aimed at achieving even higher performance, robustness, and generalization in large-scale visual applications.
Responsibilities
- You will work closely with experienced ML researchers and engineers on cutting-edge research at the intersection of computer vision and sequence modeling. Your work will include:Designing and experimenting with new ML architectures for structured visual data.
- Evaluating alternative modeling paradigms (e.g., encoder-decoder, hybrid Transformer models, sequence-based representations).
- Investigating techniques for improving robustness, generalization, and multi-view reasoning.
- Running systematic experiments, ablations, and error analyses to validate research hypotheses.
- This project provides opportunities for novel model design, extensive experimentation, and scholarly research. You will contribute to long-term innovation in our technology, with potential real-world impact for millions of users. An ideal position for experienced master's students, PhD collaborations, or candidates preparing fo