Fellowship - Deep Learning for Calibration

Tangram Vision

Remote Remote US Time Zones

Full time

Trucks VC portfolio

May 9

Tangram Vision is a mission-driven company devoted to empowering all developers to be perception creators. It’s a tall order, but our whole team is experienced in computer vision and has learned what works (and what doesn’t) in the real world. We’re honing that experience into a one-of-a-kind premium perception offering.

This is a full-time fellowship opportunity with a duration of 6 months to 1 year.

This fellowship centered around deep learning for calibration is all about one goal: moving "away from the checkerboard". No one likes waving fiducials, targets, and boards around a scene to get the right calibration data. What if we didn't have to? What if these features could be learned, or derived? How can we do this in 2D? In 3D? All of these questions can and will be explored.

This fellowship is ideally suited for those who are looking to apply their academic experience with deep learning and perception in an applied setting. It will push the candidate towards new and novel research in calibration and perception, while will also pushing us at Tangram to adopt deep learning methods into our platform. We sincerely hope that there will be enough established to generate a publication from this work, or at the very least get started towards that goal. 

Starting Salary Range: $40,000 USD for 6 mos. $80,000 USD for 1 year.

Research Milestones

1. Creation of learned 2D-2D correspondences:

  • Possible inputs: IR data, RGB data, grayscale data.
  • Deriving metric information from these datasets is not necessary; this is a non-goal.
  • Datasets will be of unknown size, and could be any combination of the above input types.
  • Features are specifically designed to fit into a larger geometric calibration pipeline.

2. Creation of learned 2D-3D correspondences:

  • Possible inputs: All 2D data types described in goal 1, along with corresponding LiDAR and depth (2.5D) datasets
  • Metric information can be derived from the 3D data.
  • Datasets will be of unknown size, and could be any combination of the above input types.
  • Features are specifically designed to fit into a larger geometric calibration pipeline.

3. Rectification and registration of 2D and 3D data based off of derived feature pipeline.

We do not expect candidates to solve, or even approach, all three of these goals during their fellowship period. However, Tangram is committed to understanding long-term calibration challenges through the lens of learned neural networks, and this research will be our first steps in that direction.

Timeline

First Month: Literature Review

The first several weeks of this fellowship will be strictly about gathering information.

  • What prior art does the candidate see being relevant to your work?
  • What platforms and open-source solutions will be helpful in the candidate's implementation?
  • What unknowns do they need to understand before diving into the work?
  • Are there existing benchmarks or datasets that are particularly useful?

This deep dive will form the foundation of the fellowship work. This should be openly documented and shared with the team at large. During this time, if needed, we can further refine the goals of the fellowship.

Every Two Weeks: Progress and Next Steps

Every other Monday, plan to present progress to the team.

This update will primarily be written. We ask that candidates document their work, experiments, and failures thoroughly as they go. This can be summarized and sent out to everyone on the engineering team before the meeting.

Examples of Things You’ll Work On...

  • Develop performant deep learning processes for both cloud and local machines
  • Develop unit and integration tests for all production-critical code
  • Participate and contribute to intensive research-backed product proposals
  • Write and contribute to our internal and user-facing product documentation
  • Contribute to the Tangram Visions blog with relevant, technical content for a robotics and sensor savvy audience
  • Be open to networking/speaker opportunities on behalf of the company:
  • Presentations at local meetups
  • Communications inside and outside of Tangram on progress and projects
  • Communicate and check-in with each other. Every proposal, piece of documentation, and prototype is meant to be open, malleable, and collaborative. This lets us do our best work while preserving the best aspects of asynchronous work - focus and time. Even so, live communications are encouraged even if its just to check in on someone’s day
  • Do your part in keeping our operations running smoothly - submitting PTO requests, observing travel budgets, etc.

You’ll be a successful candidate if...

  • You take the time to evaluate technology options and tradeoffs, and feel certain in your decision-making after evaluation
  • You care about contributing to product direction, planning, roadmaps, and postmortems
  • You have a willingness to voice your opinions and have pushed for better calibration procedures at your previous place of work or study, or have developed improved calibration procedures yourself
  • You have strong verbal and written communications skills - you’re able to explain big ideas without technical jargon
  • You have exceptional time management, including the ability to manage multiple tasks and projects at once
  • You love to learn and can implement what you’ve absorbed, whether it’s through reading textbooks and/or research papers
  • You have a strong sense of teamwork - you enjoy building positive relationships with colleagues and stakeholders, and a willingness to help others
  • You believe in inclusivity when working with colleagues with different backgrounds, experiences, and perspectives.

Must-Have Technical Qualifications...

In our experience, people with the below experiences would thrive in this role. Our tech stack is shown in parentheses, where applicable; experience with a specific tool is not required for this role.

  • At least one year of academic or professional experience with calibration procedures in perception
  • Comfortable working on Linux and using the command-line
  • Demonstrated experience with...
  • TensorFlow, Caffe, PyTorch, Keras, or another ML framework
  • Common deep learning methodologies and practices
  • Common sensor datatypes, e.g. image formats, point clouds, gyroscopic data, etc.
  • Proficiency with git. You have experience with informative commit messages and keeping a clean git history

Benefits and other Perks...

We’re a startup, aren't there supposed to be pool tables and stuff? We have an ever-growing list of items that we’re looking to provide, here’s what we currently offer.

  • We value your health and well-being: Tangram Vision covers up to $400 a month of medical plan costs (chances are we have a plan that will be 100% covered for you), and 100% of dental, vision, and ancillary plans (i.e. life insurance, short-term and long-term disability).
  • Employees who choose a medical plan with a health savings account (HSA) also receive a $2,000/year HSA contribution
  • Staff have an unlimited PTO balance, with a recommended 4 weeks off per year (1 week per quarter).
  • Remote First, Fully Connected: Tangram Vision is a fully remote company. We have no corporate office, and our first five employees span four timezones!
  • Company retreats every three to four months in fun, unique locations.
  • Weekly team-wide calls keep us connected and grounded
  • Open virtual co-working comms every afternoon
  • We understand the importance of financial planning: We offer competitive salaries for the market.

How to Apply...

  • Submit your resume and cover letter
  • In your cover letter please include how you heard about this role and why you’re interested in this position
  • Include any GitHub projects, portfolios, or deployed projects that support your experience

No phone calls please. Tangram Vision will contact qualified candidates directly.

Tangram Vision provides equal employment opportunities (EEO) to all employees and applicants for employment without regard to race, color, religion, national origin, age, disability, genetics, sexual orientation, gender identity, gender expression or veteran status. We are committed to establishing a diverse and inclusive workforce and encourage people from all backgrounds, experiences, perspectives, and abilities to apply. We do not discriminate in our hiring practices and in order to build the strongest possible workforce, actively seek a diverse applicant pool.

About Tangram Vision

At Tangram Vision, we’re building a platform of products that help robotics leaders, engineers, and fleet managers understand their perception systems inside and out at scale. By taking the pain out of working with sensors, we enable customers to focus on building solutions in their problem domains, which range from farming and warehouse logistics to autonomous vehicles and construction.

We’re Decision-Oriented

Trust is the backbone of our team. To build trust, we seek to better understand one another through our conversations, our retreats, our User Manuals, and our written viewpoints.

Trust in your choices, and the choices of others, allows Tangram to be a place of context, not control. Information is created deliberately and shared broadly by default (which means we write a lot of documentation). This has helped keep velocity up at a breakneck speed from day 1.

Dissent and Diversity

Though everyone is decision-oriented, we do so through seeking dissent. The more concerns we hear, and the more varied those concerns are, the better our outcomes become. In this way, diversity and inclusion are more than just good civic practices; they’re vital to Tangram’s success. As a startup, we are committed to keeping diversity and inclusion as a principal pillar behind all we do and how we grow.

Apply for this position Back to job

You must be logged in to to apply to this job.

Apply

Your application has been successfully submitted.

Please fix the errors below and resubmit.

Something went wrong. Please try again later or contact us.

Personal Information

Profile

View resume

Details

Tangram Vision

Perception DevOps

{{notification.msg}}