WHAT THE FUCK OF THE M2DET!!!!!

[WIKI2020]

I SET m2det512_vgg :
train_cfg = dict(
cuda = True,
warmup = 5,
per_batch_size = 32,
lr = [0.004, 0.002, 0.0004, 0.00004, 0.000004],
gamma = 0.1,
end_lr = 1e-6,
step_lr = dict(
COCO = [50, 100, 150, 200, 250],
VOC = [50, 100, 150, 200, 250], # unsolve
),
print_epochs = 10,
num_workers= 8,
)

and keep_per_class=1，score_threshold=0.9，voc0712.py and Other parameters should have been set correctly

train with nvidia v100 gpu with 3000 picture and 3 class, no problem

but !!!!!!!!!! run demo.py there are also wrong boxes in the picture[My example is that a picture must have three frames(3 class), each of which has a probability of more than 90%]

how can i do????

fuck!! yolov3 is much better

[WIKI2020]

I use the pre-training model vgg16_reducedfc.pth

[Ikhwansong]

Hey just calm down please.
step by step, you can do it. cheer up !