I tried just naively replicating the amp and horovod settings from train_ssd.py to train_center_net.py.
Here’s my modification attempt for train_center_net.py:
"""Train CenterNet"""
import argparse
import os
import logging
import warnings
import time
import numpy as np
import mxnet as mx
from mxnet import nd
from mxnet import gluon
from mxnet import autograd
import gluoncv as gcv
gcv.utils.check_version('0.8.0')
from gluoncv import data as gdata
from gluoncv import utils as gutils
from gluoncv.model_zoo import get_model
from gluoncv.data.batchify import Tuple, Stack, Pad
from gluoncv.data.transforms.presets.center_net import CenterNetDefaultTrainTransform
from gluoncv.data.transforms.presets.center_net import CenterNetDefaultValTransform, get_post_transform
from gluoncv.utils.metrics.voc_detection import VOC07MApMetric
from gluoncv.utils.metrics.coco_detection import COCODetectionMetric
from gluoncv.utils.metrics.accuracy import Accuracy
from gluoncv.utils import LRScheduler, LRSequential
from mxnet.contrib import amp
try:
import horovod.mxnet as hvd
except ImportError:
hvd = None
try:
from nvidia.dali.plugin.mxnet import DALIGenericIterator
dali_found = True
except ImportError:
dali_found = False
def parse_args():
parser = argparse.ArgumentParser(description='Train CenterNet networks.')
parser.add_argument('--network', type=str, default='resnet18_v1b',
help="Base network name which serves as feature extraction base.")
parser.add_argument('--data-shape', type=int, default=512,
help="Input data shape, use 300, 512.")
parser.add_argument('--batch-size', type=int, default=32,
help='Training mini-batch size')
parser.add_argument('--dataset', type=str, default='voc',
help='Training dataset. Now support voc.')
parser.add_argument('--dataset-root', type=str, default='~/.mxnet/datasets/',
help='Path of the directory where the dataset is located.')
parser.add_argument('--num-workers', '-j', dest='num_workers', type=int,
default=4, help='Number of data workers, you can use larger '
'number to accelerate data loading, if you CPU and GPUs are powerful.')
parser.add_argument('--gpus', type=str, default='0',
help='Training with GPUs, you can specify 1,3 for example.')
parser.add_argument('--epochs', type=int, default=140,
help='Training epochs.')
parser.add_argument('--resume', type=str, default='',
help='Resume from previously saved parameters if not None. '
'For example, you can resume from ./ssd_xxx_0123.params')
parser.add_argument('--start-epoch', type=int, default=0,
help='Starting epoch for resuming, default is 0 for new training.'
'You can specify it to 100 for example to start from 100 epoch.')
parser.add_argument('--lr', type=float, default=1.25e-4,
help='Learning rate, default is 0.000125')
parser.add_argument('--lr-decay', type=float, default=0.1,
help='decay rate of learning rate. default is 0.1.')
parser.add_argument('--lr-decay-epoch', type=str, default='90,120',
help='epochs at which learning rate decays. default is 90,120.')
parser.add_argument('--lr-mode', type=str, default='step',
help='learning rate scheduler mode. options are step, poly and cosine.')
parser.add_argument('--warmup-lr', type=float, default=0.0,
help='starting warmup learning rate. default is 0.0.')
parser.add_argument('--warmup-epochs', type=int, default=0,
help='number of warmup epochs.')
parser.add_argument('--momentum', type=float, default=0.9,
help='SGD momentum, default is 0.9')
parser.add_argument('--wd', type=float, default=0.0001,
help='Weight decay, default is 1e-4')
parser.add_argument('--log-interval', type=int, default=100,
help='Logging mini-batch interval. Default is 100.')
parser.add_argument('--num-samples', type=int, default=-1,
help='Training images. Use -1 to automatically get the number.')
parser.add_argument('--save-prefix', type=str, default='',
help='Saving parameter prefix')
parser.add_argument('--save-interval', type=int, default=10,
help='Saving parameters epoch interval, best model will always be saved.')
parser.add_argument('--val-interval', type=int, default=1,
help='Epoch interval for validation, increase the number will reduce the '
'training time if validation is slow.')
parser.add_argument('--seed', type=int, default=233,
help='Random seed to be fixed.')
parser.add_argument('--wh-weight', type=float, default=0.1,
help='Loss weight for width/height')
parser.add_argument('--center-reg-weight', type=float, default=1.0,
help='Center regression loss weight')
parser.add_argument('--flip-validation', action='store_true',
help='flip data augmentation in validation.')
parser.add_argument('--dali', action='store_true',
help='Use DALI for data loading and data preprocessing in training. '
'Currently supports only COCO.')
parser.add_argument('--amp', action='store_true',
help='Use MXNet AMP for mixed precision training.')
parser.add_argument('--horovod', action='store_true',
help='Use MXNet Horovod for distributed training. Must be run with OpenMPI. '
'--gpus is ignored when using --horovod.')
args = parser.parse_args()
if args.horovod:
assert hvd, "You are trying to use horovod support but it's not installed"
return args
def get_dataset(dataset, args):
if dataset.lower() == 'voc':
train_dataset = gdata.VOCDetection(
splits=[(2007, 'trainval'), (2012, 'trainval')])
val_dataset = gdata.VOCDetection(
splits=[(2007, 'test')])
val_metric = VOC07MApMetric(iou_thresh=0.5, class_names=val_dataset.classes)
elif dataset.lower() == 'coco':
train_dataset = gdata.COCODetection(root=args.dataset_root + "/coco", splits='instances_train2017')
val_dataset = gdata.COCODetection(root=args.dataset_root + "/coco", splits='instances_val2017', skip_empty=False)
val_metric = COCODetectionMetric(
val_dataset, args.save_prefix + '_eval', cleanup=True,
data_shape=(args.data_shape, args.data_shape), post_affine=get_post_transform)
# coco validation is slow, consider increase the validation interval
if args.val_interval == 1:
args.val_interval = 10
else:
raise NotImplementedError('Dataset: {} not implemented.'.format(dataset))
if args.num_samples < 0:
args.num_samples = len(train_dataset)
return train_dataset, val_dataset, val_metric
def get_dataloader(net, train_dataset, val_dataset, data_shape, batch_size, num_workers, ctx):
"""Get dataloader."""
width, height = data_shape, data_shape
num_class = len(train_dataset.classes)
batchify_fn = Tuple([Stack() for _ in range(6)]) # stack image, cls_targets, box_targets
train_loader = gluon.data.DataLoader(
train_dataset.transform(CenterNetDefaultTrainTransform(
width, height, num_class=num_class, scale_factor=net.scale)),
batch_size, True, batchify_fn=batchify_fn, last_batch='rollover', num_workers=num_workers)
val_batchify_fn = Tuple(Stack(), Pad(pad_val=-1))
val_loader = gluon.data.DataLoader(
val_dataset.transform(CenterNetDefaultValTransform(width, height)),
batch_size, False, batchify_fn=val_batchify_fn, last_batch='keep', num_workers=num_workers)
return train_loader, val_loader
def save_params(net, best_map, current_map, epoch, save_interval, prefix):
current_map = float(current_map)
if current_map > best_map[0]:
best_map[0] = current_map
net.save_parameters('{:s}_best.params'.format(prefix, epoch, current_map))
with open(prefix+'_best_map.log', 'a') as f:
f.write('{:04d}:\t{:.4f}\n'.format(epoch, current_map))
if save_interval and epoch % save_interval == 0:
net.save_parameters('{:s}_{:04d}_{:.4f}.params'.format(prefix, epoch, current_map))
def validate(net, val_data, ctx, eval_metric, flip_test=False):
"""Test on validation dataset."""
eval_metric.reset()
net.flip_test = flip_test
mx.nd.waitall()
net.hybridize()
for batch in val_data:
data = gluon.utils.split_and_load(batch[0], ctx_list=ctx, batch_axis=0, even_split=False)
label = gluon.utils.split_and_load(batch[1], ctx_list=ctx, batch_axis=0, even_split=False)
det_bboxes = []
det_ids = []
det_scores = []
gt_bboxes = []
gt_ids = []
gt_difficults = []
for x, y in zip(data, label):
# get prediction results
ids, scores, bboxes = net(x)
det_ids.append(ids)
det_scores.append(scores)
# clip to image size
det_bboxes.append(bboxes.clip(0, batch[0].shape[2]))
# split ground truths
gt_ids.append(y.slice_axis(axis=-1, begin=4, end=5))
gt_bboxes.append(y.slice_axis(axis=-1, begin=0, end=4))
gt_difficults.append(y.slice_axis(axis=-1, begin=5, end=6) if y.shape[-1] > 5 else None)
# update metric
eval_metric.update(det_bboxes, det_ids, det_scores, gt_bboxes, gt_ids, gt_difficults)
return eval_metric.get()
def train(net, train_data, val_data, eval_metric, ctx, args):
"""Training pipeline"""
net.collect_params().reset_ctx(ctx)
# lr decay policy
lr_decay = float(args.lr_decay)
lr_steps = sorted([int(ls) for ls in args.lr_decay_epoch.split(',') if ls.strip()])
lr_decay_epoch = [e - args.warmup_epochs for e in lr_steps]
num_batches = args.num_samples // args.batch_size
lr_scheduler = LRSequential([
LRScheduler('linear', base_lr=0, target_lr=args.lr,
nepochs=args.warmup_epochs, iters_per_epoch=num_batches),
LRScheduler(args.lr_mode, base_lr=args.lr,
nepochs=args.epochs - args.warmup_epochs,
iters_per_epoch=num_batches,
step_epoch=lr_decay_epoch,
step_factor=args.lr_decay, power=2),
])
for k, v in net.collect_params('.*bias').items():
v.wd_mult = 0.0
if args.horovod:
hvd.broadcast_parameters(net.collect_params(), root_rank=0)
trainer = hvd.DistributedTrainer(
net.collect_params(), 'adam',
{'learning_rate': args.lr, 'wd': args.wd, 'lr_scheduler': lr_scheduler})
else:
trainer = gluon.Trainer(
net.collect_params(), 'adam',
{'learning_rate': args.lr, 'wd': args.wd, 'lr_scheduler': lr_scheduler},
update_on_kvstore=(False if args.amp else None))
# https://mxnet.apache.org/versions/1.7/api/python/docs/tutorials/performance/backend/amp.html#Training-with-AMP
if args.amp:
amp.init_trainer(trainer)
heatmap_loss = gcv.loss.HeatmapFocalLoss(from_logits=True)
wh_loss = gcv.loss.MaskedL1Loss(weight=args.wh_weight)
center_reg_loss = gcv.loss.MaskedL1Loss(weight=args.center_reg_weight)
heatmap_loss_metric = mx.metric.Loss('HeatmapFocal')
wh_metric = mx.metric.Loss('WHL1')
center_reg_metric = mx.metric.Loss('CenterRegL1')
# set up logger
logging.basicConfig()
logger = logging.getLogger()
logger.setLevel(logging.INFO)
log_file_path = args.save_prefix + '_train.log'
log_dir = os.path.dirname(log_file_path)
if log_dir and not os.path.exists(log_dir):
os.makedirs(log_dir)
fh = logging.FileHandler(log_file_path)
logger.addHandler(fh)
logger.info(args)
logger.info('Start training from [Epoch {}]'.format(args.start_epoch))
best_map = [0]
for epoch in range(args.start_epoch, args.epochs):
wh_metric.reset()
center_reg_metric.reset()
tic = time.time()
btic = time.time()
net.hybridize()
for i, batch in enumerate(train_data):
split_data = [gluon.utils.split_and_load(batch[ind], ctx_list=ctx, batch_axis=0) for ind in range(6)]
data, heatmap_targets, wh_targets, wh_masks, center_reg_targets, center_reg_masks = split_data
batch_size = args.batch_size
with autograd.record():
sum_losses = []
heatmap_losses = []
wh_losses = []
center_reg_losses = []
wh_preds = []
center_reg_preds = []
for x, heatmap_target, wh_target, wh_mask, center_reg_target, center_reg_mask in zip(*split_data):
heatmap_pred, wh_pred, center_reg_pred = net(x)
wh_preds.append(wh_pred)
center_reg_preds.append(center_reg_pred)
wh_losses.append(wh_loss(wh_pred, wh_target, wh_mask))
center_reg_losses.append(center_reg_loss(center_reg_pred, center_reg_target, center_reg_mask))
heatmap_losses.append(heatmap_loss(heatmap_pred, heatmap_target))
curr_loss = heatmap_losses[-1]+ wh_losses[-1] + center_reg_losses[-1]
sum_losses.append(curr_loss)
if args.amp:
with amp.scale_loss(sum_losses, trainer) as scaled_loss:
autograd.backward(scaled_loss)
else:
autograd.backward(sum_losses)
trainer.step(len(sum_losses)) # step with # gpus
if (not args.horovod or hvd.rank() == 0):
local_batch_size = int(args.batch_size // (hvd.size() if args.horovod else 1))
heatmap_loss_metric.update(0, [l * local_batch_size for l in heatmap_losses])
wh_metric.update(0, [l * local_batch_size for l in wh_losses])
center_reg_metric.update(0, [l * local_batch_size for l in center_reg_losses])
if args.log_interval and not (i + 1) % args.log_interval:
name2, loss2 = wh_metric.get()
name3, loss3 = center_reg_metric.get()
name4, loss4 = heatmap_loss_metric.get()
logger.info('[Epoch {}][Batch {}], Speed: {:.3f} samples/sec, LR={}, {}={:.3f}, {}={:.3f}, {}={:.3f}'.format(
epoch, i, batch_size/(time.time()-btic), trainer.learning_rate, name2, loss2, name3, loss3, name4, loss4))
btic = time.time()
if (not args.horovod or hvd.rank() == 0):
name2, loss2 = wh_metric.get()
name3, loss3 = center_reg_metric.get()
name4, loss4 = heatmap_loss_metric.get()
logger.info('[Epoch {}] Training cost: {:.3f}, {}={:.3f}, {}={:.3f}, {}={:.3f}'.format(
epoch, (time.time()-tic), name2, loss2, name3, loss3, name4, loss4))
if (epoch % args.val_interval == 0) or (args.save_interval and epoch % args.save_interval == 0) or (epoch == args.epochs - 1):
# consider reduce the frequency of validation to save time
map_name, mean_ap = validate(net, val_data, ctx, eval_metric, flip_test=args.flip_validation)
val_msg = '\n'.join(['{}={}'.format(k, v) for k, v in zip(map_name, mean_ap)])
logger.info('[Epoch {}] Validation: \n{}'.format(epoch, val_msg))
current_map = float(mean_ap[-1])
else:
current_map = 0.
save_params(net, best_map, current_map, epoch, args.save_interval, args.save_prefix)
if __name__ == '__main__':
args = parse_args()
if args.amp:
amp.init()
if args.horovod:
hvd.init()
# fix seed for mxnet, numpy and python builtin random generator.
gutils.random.seed(args.seed)
# training contexts
if args.horovod:
ctx = [mx.gpu(hvd.local_rank())]
else:
ctx = [mx.gpu(int(i)) for i in args.gpus.split(',') if i.strip()]
ctx = ctx if ctx else [mx.cpu()]
# network
net_name = '_'.join(('center_net', args.network, args.dataset))
args.save_prefix += net_name
net = get_model(net_name, pretrained_base=True, norm_layer=gluon.nn.BatchNorm)
async_net = net
if args.resume.strip():
net.load_parameters(args.resume.strip())
async_net.load_parameters(args.resume.strip())
else:
with warnings.catch_warnings(record=True) as w:
warnings.simplefilter("always")
net.initialize()
async_net.initialize
# needed for net to be first gpu when using AMP
net.collect_params().reset_ctx(ctx[0])
# training data
if args.dali:
if not dali_found:
raise SystemExit("DALI not found, please check if you installed it correctly.")
devices = [int(i) for i in args.gpus.split(',') if i.strip()]
train_dataset, val_dataset, eval_metric = get_dali_dataset(args.dataset, devices, args)
train_data, val_data = get_dali_dataloader(
async_net, train_dataset, val_dataset, args.data_shape, args.batch_size, args.num_workers,
devices, ctx[0], args.horovod)
else:
train_dataset, val_dataset, eval_metric = get_dataset(args.dataset, args)
batch_size = (args.batch_size // hvd.size()) if args.horovod else args.batch_size
train_data, val_data = get_dataloader(
async_net, train_dataset, val_dataset, args.data_shape, batch_size, args.num_workers, ctx[0])
# training
train(net, train_data, val_data, eval_metric, ctx, args)
However I got the following error for AMP:
terminate called after throwing an instance of ‘dmlc::Error’
what(): [23:02:20] src/operator/tensor/./…/./…/common/…/operator/nn/mkldnn/mkldnn_base-inl.h:246: unknown type for MKLDNN :2
And quite a long error from Horovod that starts with:
[1,3]:Signal: Bus error (7)
[1,3]:Signal code: Non-existant physical address (2)
I assume the workings of both AMP and Horovod are more complex than I realized and can’t simply be copied from the SSD script. Does anyone know how to go about making either of these work for this model? If anyone has insight into these issues I’d really appreciate it.