results_log

Using pretrained coco model


Training

Version 2
Images: Training: 777, Test: 326
Annotations: Training: 12703, Test: 5327
Trained for 8 epochs
Results on test set
Biker: 0.4879
Car:0.9528
Bus: 0.8571
Cart:0.1957
Skater:0.0170
Pedestrian: 0.500
Weighted: 0.5048

Version 3 - More data
Images: Training: 2246, Test: 1032
Annotations: Training: 30566, Test: 14195
Trained for 25 epochs
Results on test set
Biker: 0.4825
Car:0.9576
Bus: 0.8896
Cart:0.1974
Skater:0.0848
Pedestrian: 0.6428
Weighted: 0.5964


Version 4 - Changing anchor sizes

Images: Training: 2246, Test: 1032
Annotations: Training: 30566, Test: 14195
Trained for 17 epochs
Results on test set
Biker: 0.4862
Car:0.9363
Bus: 0.7892
Cart:0.1510
Skater:0.0807
Pedestrian: 0.7059
Weighted: 0.6376


Version 5 - new test set and feeeze backbone

Biker: 0.5432
Car:0.9787
Bus: 0.4318
Cart:0.0270
Skater:0.0913
Pedestrian: 0.6255
Weighted: 0.5999

1799 instances of class Biker with average precision: 0.6081
162 instances of class Car with average precision: 0.9802
34 instances of class Bus with average precision: 0.8109
40 instances of class Cart with average precision: 0.1062
53 instances of class Skater with average precision: 0.2389
3766 instances of class Pedestrian with average precision: 0.6882
mAP using the weighted average of precisions among classes: 0.6643

With very first weight
1799 instances of class Biker with average precision: 0.6209
162 instances of class Car with average precision: 0.9832
34 instances of class Bus with average precision: 0.8373
40 instances of class Cart with average precision: 0.2057
53 instances of class Skater with average precision: 0.2736
3766 instances of class Pedestrian with average precision: 0.6825
mAP using the weighted average of precisions among classes: 0.6658