def run(self):
ae_res = []
stability_res = []
# Cross Validation loop
for i in range(len(self.folds)):
# get first fold #i info
fold = self.folds[i]
# set train, validation and test sets for this fold iteration
self.args.trainfiles = fold.train
self.args.valfile = fold.validation
self.args.testfile = fold.test
# all training logic is inside Worker (core/worker.py)
worker = Worker(i, self.conf, self.data_conf, self.cache, self.args,
verbose=self.verbose, inference=self.inference)
res, _ = worker.run()
# print results for each fold
if self.verbose:
partial_res = summary_angular_errors([r.error for r in res])
self._print_results(partial_res, 'fold '+str(i))
# accumulate results into lists
ae_res += res
# summary_angular_errors: computes mean, median, best 25%, etc...
results = summary_angular_errors([r.error for r in ae_res])
if self.verbose:
self._print_results(results, 'total')
return results
def main():
args = parser.parse_args()
if args.seed is not None:
random.seed(args.seed)
torch.manual_seed(args.seed)
cudnn.deterministic = True
warnings.warn('You have chosen to seed training. '
'This will turn on the CUDNN deterministic setting, '
'which can slow down your training considerably! '
'You may see unexpected behavior when restarting '
'from checkpoints.')
# load configuration file: epochs, loss function, etc... for this experiment
with open(args.configurationfile, 'r') as f:
conf = json.load(f)
# load datapath file: paths specific to the current machine
with open(args.datapath, 'r') as f:
data_conf = json.load(f)
# remove previous results
output_dir = os.path.join(args.outputfolder, args.dataset, args.subdataset,
conf['name'])
if not args.evaluate and not args.resume:
if os.path.exists(output_dir):
shutil.rmtree(output_dir)
# create output folder
os.makedirs(output_dir, exist_ok=True)
args.outputfolder = output_dir
# copy configuration file to output folder
shutil.copy(
args.configurationfile,
os.path.join(output_dir, os.path.basename(args.configurationfile)))
# we overwrite the stdout and stderr (standard output and error) to
# files in the output directory
sys.stdout = PrintLogger(os.path.join(output_dir, 'stdout.txt'),
sys.stdout)
sys.stderr = PrintLogger(os.path.join(output_dir, 'stderr.txt'),
sys.stderr)
fold = 0 # no folds, but we always use fold #0 for these experiments
cache = CacheManager(conf)
worker = Worker(fold, conf, data_conf, cache, args)
res, _ = worker.run()
# some datasets have no validation GT
if len(res) > 0:
# print angular errors statistics (mean, median, etc...)
generate_results(res, 'test')
def main():
args = parser.parse_args()
# load configuration file for this experiment
with open(args.configurationfile, 'r') as f:
conf = json.load(f)
# load datapath file: paths specific to the current machine
with open(args.datapath, 'r') as f:
data_conf = json.load(f)
# create output folder
os.makedirs(args.outputfolder, exist_ok=True)
# copy configuration file to output folder
shutil.copy(
args.configurationfile,
os.path.join(args.outputfolder,
os.path.basename(args.configurationfile)))
# we overwrite the stdout and stderr (standard output and error) to
# files in the output directory
sys.stdout = PrintLogger(os.path.join(args.outputfolder, 'stdout.txt'),
sys.stdout)
sys.stderr = PrintLogger(os.path.join(args.outputfolder, 'stderr.txt'),
sys.stderr)
# used in core/worker.py to determine what to do
args.evaluate = True
args.resume = False
args.valfile = None
args.trainfiles = None
# init the cache manager: this caches images from the dataset
# to avoid reading them more than once
cache = CacheManager(conf, no_cache=True)
fold = 0 # no folds, but we always use fold #0 for these experiments
worker = Worker(fold, conf, data_conf, cache, args, inference=True)
res, _ = worker.run()
# print angular errors statistics (mean, median, etc...)
generate_results(res)
def main():
# Initialize worker
worker = Worker()
# Run worker
worker.run()
def runWorker(url):
log.info("Worker URL: " + url)
worker = Worker()
worker.run(url)
def runWorker(url):
log.info("Worker URL: " + url)
worker = Worker()
worker.run(url)
