def main(client_num,
model,
C=1.0,
lr=0.1,
epoch=1,
aggregator=FedAvg(),
iid=True,
client_eval=False):
logger = Logger('output/{}'.format(
datetime.now().strftime('%y%m%d%H%M%S')),
level='debug')
# 读取配置
cfg.merge_from_file('config/{}.yaml'.format(model))
if C == 0:
cfg.SOLVER.MAX_EPOCH = 2000
cfg.freeze()
# 加载数据
transforms_train = transforms.Compose([
transforms.Resize(cfg.MODEL.BACKBONE.INPUT_RESOLUTION),
transforms.ToTensor(),
transforms.Normalize((0.1307, ), (0.3081, ))
])
train_set = torchvision.datasets.MNIST(root='data',
train=True,
download=True,
transform=transforms_train)
# 分配数据
if iid:
sample_idx = iid_sampler(train_set, client_num)
else:
sample_idx = non_iid_sampler(train_set, client_num)
# 初始化联邦
s = Server(config=cfg,
dataset_name='MNIST',
aggregator=aggregator,
logger=logger)
def flip(tensor):
return (tensor + 2) % 10
for i, sample in enumerate(sample_idx):
if i < client_num - 10:
c = Client(i, train_set, sample_idx[i], logger=logger)
else:
c = LabelFlippingClient(i,
train_set,
sample_idx[i],
label_map=flip,
logger=logger)
s.append_clients(c)
s.train(C=C, epoch_num=epoch, lr=lr, local_eval=client_eval)
cfg.defrost()
def main():
''' Main function '''
if cfg.SYSTEM.UPDATE_CFG:
cfg.merge_from_file(cfg.SYSTEM.CFG_FILE)
cfg.freeze() # 冻结参数
print(cfg)
vis.log('configs:\n {}'.format(cfg.clone()))
# LOAD DATA
from data import PascalBox
dataset = PascalBox()
# training_data, validation_data = prepare_stanford40(cfg.clone())
# training_data, validation_data = prepare_pascal(cfg.clone())
training_data = DataLoader(dataset, batch_size=46, shuffle=True)
validation_data = DataLoader(dataset, batch_size=46)
from lib.attention_reid.attention_reid import attentionReid
# LOAD MODEL
model = attentionReid()
# model = model.multi_two(pretrained=True)
for params in model.backbone.parameters():
params.require_grad = False
model = model.cuda()
# ------------- optim -----------------#
# optimizer = optim.SGD([
# {'params': model.backbone.parameters(),'lr':1e-4},
# {'params': model.attention.parameters(), 'lr': 1e-3},
# {'params': model.GCN.parameters(),'lr':1e-3}
# ], lr=1e-3,)
optimizer = optim.SGD(model.parameters(), lr=1e-4)
scheduler = ReduceLROnPlateau(optimizer, mode='min', factor=cfg.OPTIM.LR_DECAY, patience=1,
verbose=True, threshold=1e-4, threshold_mode='rel',
cooldown=1, min_lr=0.00005, eps=1e-8)
# ------------- train ----------------#
train(model, training_data, validation_data, optimizer, scheduler, cfg.clone())
import torch
import torch.nn.functional as F
import torch.optim as optim
from torch.utils.data import DataLoader
from torch.autograd import Variable as V
from torchnet import meter
from config.config import cfg
from util.visualize import Visualizer
from util.show_masked_image import show_masked_image
from mmcv.runner import save_checkpoint, load_checkpoint
import cv2
from util.show_masked_image import tensor_to_np
import numpy as np
#cfg.merge_from_file("config/un_att_pascal_0001.yaml")
cfg.freeze() # 冻结参数
vis = Visualizer("newvis", port=8097)
AP = meter.APMeter()
mAP = meter.mAPMeter()
top3 = meter.ClassErrorMeter(topk=[1, 3, 5], accuracy=True)
Loss_meter = meter.AverageValueMeter()
os.environ["CUDA_VISIBLE_DEVICES"] = "0,1,2"
num = 30
def visualize_func(result):
pass
def inverse_normalize(img):
import json
from config.config import cfg
cfg.merge_from_file('./config/config.yaml')
cfg.freeze()
print(json.dumps(cfg, indent=4))