The learing rate is always 0

[zx-code123]

@potterhsu When i train with PASCAL VOC 2007 ,learning rate is always 0 like this.

[potterhsu]

Could I have the Python and PyTorch versions you’re running on?

[zx-code123]

Could I have the Python and PyTorch versions you’re running on?

python version:3.6.7 pytorch version:1.1.0 CUDA Version: 10.0

[potterhsu]

This project does not support PyTorch 1.1.0 yet, you may try following codes instead in lr_scheduler.py

from bisect import bisect_right
from typing import List

from torch.optim import Optimizer
from torch.optim.lr_scheduler import _LRScheduler

class WarmUpMultiStepLR(_LRScheduler):
    def __init__(self, optimizer: Optimizer, milestones: List[int], gamma: float = 0.1,
                 factor: float = 0.3333, num_iters: int = 500, last_epoch: int = -1):
        self.milestones = milestones
        self.gamma = gamma
        self.factor = factor
        self.num_iters = num_iters
        super().__init__(optimizer, last_epoch)

    def get_lr(self) -> List[float]:
        if self.last_epoch < self.num_iters:
            alpha = self.last_epoch / self.num_iters
            factor = (1 - self.factor) * alpha + self.factor
        else:
            factor = 1

        decay_power = bisect_right(self.milestones, self.last_epoch)
        return [base_lr * factor * self.gamma ** decay_power
                for base_lr in self.base_lrs]

[zx-code123]

It worked!!Thank you very much!