Skip to content

A tiny CNN for on-device disaster scene parsing and semantic segmentation for the lpcv.ai 2023 challenge.

Notifications You must be signed in to change notification settings

holdenb/lpcv-2023-tiny-espresso-net

Repository files navigation

LPCV 2023 - Tiny Espresso Net

Evaluation

Format

  • Input/output resolution: 512*512
  • Model Output: 14 * 512 * 512 for Channel * Height * Width. Each channel corresponds to the predicted probability for one category.

Submission

  • solution.pyz: the zipped package of solution/. zipapp should be used to compress the package.

Recommended command where solution is the name to your directory: python3.6 -m zipapp solution -p='/usr/bin/env python3.6'

Metrics

  • Accuracy: Dice Coefficient over all 14 categories. As calculated in evaluation/Accuracy.py
  • Speed: Average runtime for processing one frame (s/f). As calculated in solution/main.py

About

A tiny CNN for on-device disaster scene parsing and semantic segmentation for the lpcv.ai 2023 challenge.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published