Segmentation experiments 2/3/21

Created on May 26|Last edited on July 29
Comment
﻿
FindingsRan a sweep with 84 runs over a 7 hyper-parameters.
Note: top models are AlexNet!
﻿
0.0000.0010.0020.0030.0040.0050.0060.0070.008lea...0102030405060708090100110120num...alexnetresnet18resnet34enc...5.05.25.45.65.86.06.26.46.66.87.07.27.47.67.88.0bat...0.000.010.020.030.040.050.060.070.080.090.10wei...1.01.21.41.61.82.02.22.42.62.83.0tra...0.100.150.200.250.300.350.400.450.500.550.600.650.700.750.80iou0.0000.0020.0040.0060.0080.0100.0120.0140.0160.018hum...
Parameter importance with respect tohuman_iou
1-10
 of 13
Config parameter
Importance
Correlation
training_stages
training_stages
learning_rate
learning_rate
num_train
num_train
Runtime
Runtime
weight_decay
weight_decay
num_valid
num_valid
bn_weight_decay
bn_weight_decay
pretrained
pretrained
batch_size
batch_size
one_cycle
one_cycle
Loading...
Run set2
﻿
﻿
﻿
Run set398
﻿
These runs have very similar accuracy curves despite being distant in hyperparameter space.
﻿
﻿
Run set4
﻿
﻿
Add a comment