Spaces:
Runtime error
Runtime error
File size: 5,669 Bytes
24f9881 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 |
# MIT License
# Copyright (c) 2022 Intelligent Systems Lab Org
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
# File author: Shariq Farooq Bhat
from zoedepth.utils.misc import count_parameters, parallelize
from zoedepth.utils.config import get_config
from zoedepth.utils.arg_utils import parse_unknown
from zoedepth.trainers.builder import get_trainer
from zoedepth.models.builder import build_model
from zoedepth.data.data_mono import DepthDataLoader
import torch.utils.data.distributed
import torch.multiprocessing as mp
import torch
import numpy as np
from pprint import pprint
import argparse
import os
os.environ["PYOPENGL_PLATFORM"] = "egl"
os.environ["WANDB_START_METHOD"] = "thread"
def fix_random_seed(seed: int):
import random
import numpy
import torch
random.seed(seed)
numpy.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = True
def load_ckpt(config, model, checkpoint_dir="./checkpoints", ckpt_type="best"):
import glob
import os
from zoedepth.models.model_io import load_wts
if hasattr(config, "checkpoint"):
checkpoint = config.checkpoint
elif hasattr(config, "ckpt_pattern"):
pattern = config.ckpt_pattern
matches = glob.glob(os.path.join(
checkpoint_dir, f"*{pattern}*{ckpt_type}*"))
if not (len(matches) > 0):
raise ValueError(f"No matches found for the pattern {pattern}")
checkpoint = matches[0]
else:
return model
model = load_wts(model, checkpoint)
print("Loaded weights from {0}".format(checkpoint))
return model
def main_worker(gpu, ngpus_per_node, config):
try:
seed = config.seed if 'seed' in config and config.seed else 43
fix_random_seed(seed)
config.gpu = gpu
model = build_model(config)
model = load_ckpt(config, model)
model = parallelize(config, model)
total_params = f"{round(count_parameters(model)/1e6,2)}M"
config.total_params = total_params
print(f"Total parameters : {total_params}")
train_loader = DepthDataLoader(config, "train").data
test_loader = DepthDataLoader(config, "online_eval").data
trainer = get_trainer(config)(
config, model, train_loader, test_loader, device=config.gpu)
trainer.train()
finally:
import wandb
wandb.finish()
if __name__ == '__main__':
mp.set_start_method('forkserver')
parser = argparse.ArgumentParser()
parser.add_argument("-m", "--model", type=str, default="synunet")
parser.add_argument("-d", "--dataset", type=str, default='nyu')
parser.add_argument("--trainer", type=str, default=None)
args, unknown_args = parser.parse_known_args()
overwrite_kwargs = parse_unknown(unknown_args)
overwrite_kwargs["model"] = args.model
if args.trainer is not None:
overwrite_kwargs["trainer"] = args.trainer
config = get_config(args.model, "train", args.dataset, **overwrite_kwargs)
# git_commit()
if config.use_shared_dict:
shared_dict = mp.Manager().dict()
else:
shared_dict = None
config.shared_dict = shared_dict
config.batch_size = config.bs
config.mode = 'train'
if config.root != "." and not os.path.isdir(config.root):
os.makedirs(config.root)
try:
node_str = os.environ['SLURM_JOB_NODELIST'].replace(
'[', '').replace(']', '')
nodes = node_str.split(',')
config.world_size = len(nodes)
config.rank = int(os.environ['SLURM_PROCID'])
# config.save_dir = "/ibex/scratch/bhatsf/videodepth/checkpoints"
except KeyError as e:
# We are NOT using SLURM
config.world_size = 1
config.rank = 0
nodes = ["127.0.0.1"]
if config.distributed:
print(config.rank)
port = np.random.randint(15000, 15025)
config.dist_url = 'tcp://{}:{}'.format(nodes[0], port)
print(config.dist_url)
config.dist_backend = 'nccl'
config.gpu = None
ngpus_per_node = torch.cuda.device_count()
config.num_workers = config.workers
config.ngpus_per_node = ngpus_per_node
print("Config:")
pprint(config)
if config.distributed:
config.world_size = ngpus_per_node * config.world_size
mp.spawn(main_worker, nprocs=ngpus_per_node,
args=(ngpus_per_node, config))
else:
if ngpus_per_node == 1:
config.gpu = 0
main_worker(config.gpu, ngpus_per_node, config)
|