Yolo v1 re-implementation

Pre-training

1. pretraing for 20 convolution layers using ImageNet 2012 datasets.

• until Top 5 87% acc

Training

1. Using PASCAL 2007, 2012 datasets.

• 15496 images for training (70%)
• 6640 images for validation (30%)

2. 135 epochs, batch-size 64, momentum 0.9, decay 0.0005

3. Different learning rate for epochs

4. extensive data augmentation

Experiment

mAP 51.7 (Implement)
63.4 (In paper)

Class	aeroplane	bicycle	bird	boat	bottle	bus	car	cat	chair	cow
AP	74.8	51.7	56.9	36.6	16.0	68.0	46.3	85.1	26.7	51.7

Class	dining table	dog	horse	motorbike	person	potted plant	sheep	sofa	train	tv&monitor
AP	37.2	77.9	52.6	54.8	61.6	19.1	54.1	43.8	75.4	43.7

Test Images