Skip to content

Latest commit

 

History

History
36 lines (33 loc) · 2 KB

README.md

File metadata and controls

36 lines (33 loc) · 2 KB

VideoClassification-CNN

Lite version of the following Oxford project: Large-scale Video Classification with Convolutional Neural Networks

Choosen Sport Labels

Label ID Label Name
398 rugby
300 formula racing
361 beach volleyball
368 basketball
415 karate
277 motocross
111 kitesurfing
269 motorcycle racing
258 horse racing
200 bodybuilding

Results

  • Single Frame
  • Single Frame Context
  • Single Frame Fovea
  • Single Frame Multiresolution
  • Multi Frame Early Fusion
  • Multi Frame Late Fusion
  • Multi Frame Slow Fusion

Summary

Model Name Test Loss Test Accuracy Score Testing Time in minutes Training Time in hours
SingleResCNN-Stream_Type_None 0.96767 0.71325 0.37903 1.53495
SingleResCNN-Stream_Type_fovea 1.26127 0.58379 0.17914 1.20548
SingleResCNN-Stream_Type_context 1.17831 0.65533 0.16217 1.16977
MultiResCNN 1.12028 0.63378 0.13316 1.52294
LateFusionCNN 1.05671 0.69669 0.37975 3.01682
EarlyFusionCNN 1.08308 0.67226 0.67731 3.79941
SlowFusionCNN 1.27412 0.6623 1.42234 7.80681