training w/ full yolo cfg

Edmond Yoo

2018-09-11 7e9e289b164e9bbfb77f644b1cbfac48bc8c8408

training w/ full yolo cfg

 README.md

@@ -28,7 +28,7 @@

The second and third problems should easily be solved by further augmenting the dataset with random lighting and image skew. I'll have to think more about the first problem, though.

## Day 1
## Sept 7th, 2018
-----------------------

Added several image augmentation techniques to apply to the training set: noise, dropout, light variation, and glaring:
@@ -45,6 +45,16 @@

<img src="https://github.com/hj3yoo/darknet/blob/master/figures/1_detection_result_1.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/1_decision_result_2.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/1_decision_result_3.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/1_decision_result_4.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/1_decision_result_5.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/1_decision_result_6.jpg" width="360">

<img src="https://github.com/hj3yoo/darknet/blob/master/figures/1_learning_curve.jpg" width="360"> 
<img src="https://github.com/hj3yoo/darknet/blob/master/figures/1_learning_curve.jpg" width="640"> 

The video demo can be found here: https://www.youtube.com/watch?v=kFE_k-mWo2A&feature=youtu.be


## Sept 10th, 2018
-----------------------

I've been training a new model with a full YOLOv3 configuration (previous one used Tiny YOLOv3), and it's been taking a lot more resources:

<img src="https://github.com/hj3yoo/darknet/blob/master/figures/4_learning_curve.jpg" width="640"> 

The author of darknet did mention that full network will take significantly more training effort, so I'll just have to wait. At this rate, it should reach 50k epoch in about a week :/

 anchors.txt

@@ -1 +1 @@
118.3429,137.0897, 95.8160,181.9724, 140.4955,166.7423, 112.7262,220.6808, 129.2741,198.9876, 159.0679,197.4912, 138.1861,243.0256, 167.4683,229.0091, 165.0264,255.0887
129.1513,150.1077, 104.9181,195.1675, 145.8067,186.4670, 127.9706,231.2296, 165.9494,221.8162, 162.4676,253.0225

 cfg/yolov3-custom.cfg

@@ -1,7 +1,7 @@
[net]
# Testing
batch=8
subdivisions=4
batch=64
subdivisions=32
# Training
# batch=64
# subdivisions=16
@@ -17,7 +17,7 @@

learning_rate=0.001
burn_in=1000
max_batches = 500200
max_batches = 30000
policy=steps
steps=400000,450000
scales=.1,.1
@@ -606,7 +606,7 @@

[yolo]
mask = 6,7,8
anchors = 10,13,  16,30,  33,23,  30,61,  62,45,  59,119,  116,90,  156,198,  373,326
anchors = 129.1513,150.1077, 104.9181,195.1675, 145.8067,186.4670, 127.9706,231.2296, 165.9494,221.8162, 162.4676,253.0225
classes=1
num=9
jitter=.3
@@ -692,7 +692,7 @@

[yolo]
mask = 3,4,5
anchors = 10,13,  16,30,  33,23,  30,61,  62,45,  59,119,  116,90,  156,198,  373,326
anchors = 129.1513,150.1077, 104.9181,195.1675, 145.8067,186.4670, 127.9706,231.2296, 165.9494,221.8162, 162.4676,253.0225
classes=1
num=9
jitter=.3
@@ -779,7 +779,7 @@

[yolo]
mask = 0,1,2
anchors = 10,13,  16,30,  33,23,  30,61,  62,45,  59,119,  116,90,  156,198,  373,326
anchors = 129.1513,150.1077, 104.9181,195.1675, 145.8067,186.4670, 127.9706,231.2296, 165.9494,221.8162, 162.4676,253.0225
classes=1
num=9
jitter=.3

 darknet

Binary files differ

 figures/4_learning_curve.jpg