search

haofengac / MonoDepth-FPN-PyTorch

Single Image Depth Estimation with Feature Pyramid Network
MIT License
327 stars 69 forks source link

Lack of instructions #10

Open nikitadurasov opened 6 years ago

nikitadurasov commented 6 years ago

Now I faced a couple of problems: there are no such modules as "constants", "utils" and etc, but its not a major problem. I'm trying to run main_fpn.py with python main_fpn.py --epochs 40 --cuda --bs 4 --num_workers 3 --output_dir output_dir/, but only got NaN's as output from network and as values of losses.

Could you please write more detailed instructions how to run your code?

Usernamezhx commented 5 years ago

@NikitaDurasov do you solve the problem?

agustincl commented 5 years ago

Also here trying to "make it works"

For now: UTILS Comment from utils.net_utils import adjust_learning_rate to # from utils.net_utils import adjust_learning_rate

From https://github.com/jwyang/faster-rcnn.pytorch/blob/master/lib/model/utils/net_utils.py

Download the net_utils.py, or just define the adjust_learning_rate function

def adjust_learning_rate(optimizer, decay=0.1):
    """Sets the learning rate to the initial LR decayed by 0.5 every 20 epochs"""
    for param_group in optimizer.param_groups:
        param_group['lr'] = decay * param_group['lr']