# Any ideas on how to segment a 2D vector field?

**URL:** <https://deeptalk.lambda.ai/t/any-ideas-on-how-to-segment-a-2d-vector-field/1464>\
**Category:** Deep Learning: Getting Started\
**Created:** [February 26, 2020, 12:35am UTC](https://deeptalk.lambda.ai/t/any-ideas-on-how-to-segment-a-2d-vector-field/1464 "2020-02-26T00:35:59Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![tricostume](https://avatars.discourse-cdn.com/v4/letter/t/ce7236/32.png) [@tricostume](https://deeptalk.lambda.ai/u/tricostume)\
**Post date:** [February 26, 2020, 12:35am UTC](https://deeptalk.lambda.ai/t/any-ideas-on-how-to-segment-a-2d-vector-field/1464/1 "2020-02-26T00:35:59Z")

</div>

I am given a 2D vector field and a ROI out of which I sample a random number of n FLOAT vectors of the form (x, y, dx, dy). What could be good ideas to classify each of these vectors in any of two classes? (e.g. foreground/background in the case of optic flow). The challenge is over all with the variable input size (from 0 to n elements) for all possible given ROIs and maybe not having a given discrete structure to accommodate the vectors.