YoloV3 From Scratch On Pascal VOC

YoloV3 trained from scratch on PascalVOC dataset experiments include applying multi resolution training, data augmentation Mosaic & Mixup, with different LR's

Anant Gupta

Created on August 24|Last edited on September 11

Comment

﻿

























Run NameExperiment Description
Smart blaze 7max lr for OCP 10-4
Distinctive blaze 8max lr for OCP 10-3
Fluent puddle -8max lr for OCP 10-3 + multi-resolution training
Crimson silence 14max lr for OCP 10-3 + multi-resolution training + mosaic
﻿
Section 1﻿
train/loss, val/loss
train/loss, val/loss
0102030epoch51015202530
crimson-silence-14   train/loss
crimson-silence-14   val/loss
fluent-puddle-11   train/loss
fluent-puddle-11   val/loss
distinctive-blaze-8   train/loss
distinctive-blaze-8   val/loss
smart-blaze-7   train/loss
smart-blaze-7   val/loss
MAP
MAP
crimson-silence-14fluent-puddle-11distinctive-blaze-8smart-blaze-70.000.200.40
lr-Adam
lr-Adam
0200400600800Step0.00020.00040.00060.0008
crimson-silence-14
fluent-puddle-11
distinctive-blaze-8
smart-blaze-7
Run set4
﻿
﻿
﻿
Run set4
﻿
﻿
﻿
﻿
Run set4
﻿
﻿

Run Name	Experiment Description
Smart blaze 7	max lr for OCP 10-4
Distinctive blaze 8	max lr for OCP 10-3
Fluent puddle -8	max lr for OCP 10-3 + multi-resolution training
Crimson silence 14	max lr for OCP 10-3 + multi-resolution training + mosaic

Add a comment