Fermes / yolov3-mxnet

A minimal YOLOv3 implementation in MXNet, don't need cfg.
41 stars 12 forks source link

How could I train a model from scratch? #3

Open kvv2018 opened 6 years ago

kvv2018 commented 6 years ago

Because my dataset is very different with imagenet and coco. I also want to design a new backbone model that did not provide pre-trained weight. So, I want to train the project from scratch? Is this possible?

if args.params.endswith(".params"):
        net.load_params(args.params)
    elif args.params.endswith(".weights"):
        X = nd.uniform(shape=(1, 3, input_dim, input_dim), ctx=ctx[-1])
        net(X)
        net.load_weights(args.params, fine_tune=num_classes != 80)
    else:
        print("params {} load error!".format(args.params))
        exit()
Fermes commented 6 years ago

Just delete this part of codes to disable finetune. If you want to change network structure, edit darknet.py. And actualize this idea in http://pjreddie.com/darknet/ is more recommended. ^.^