Tiny YOLOv4 cfg modification to use sgdr

[SergioPrietoMolina]

Hi

I have been modifying yolov4-tiny-custom.cfg to use the policy sgdr when training TinyTolov4 using an aerial car dataset. The thing is I am not able to achieve more than 1500 iterations when actually the max_batches is set to 3990. If someone could help me I would really appreciate it.

The error outputted from the terminal of Google colab is the following:
`4CUDA status Error: file: ./src/network_kernels.cu : () : line: 720 : build time: Apr 21 2021 - 13:27:38

CUDA Error: an illegal memory access was encountered
CUDA Error: an illegal memory access was encountered: Success
darknet: ./src/utils.c:331: error: Assertion 0' failed.

And the modified parts are in the [net] layer:
`[net]

Training

batch=64
subdivisions=16
width=416
height=416
channels=3
momentum=0.9
decay=0.0005
angle=0
saturation = 1.5
exposure = 1.5
hue=.1
mosaic=1
mosaic_bound=1

learning_rate=0.0028
burn_in=1000
max_batches = 3990
policy=sgdr
sgdr_cycle=1000
sgdr_mult=1.3`

[SergioPrietoMolina]

Fixed by addinging <steps=3200,3600> below the policy.