def evaluate(self, category, train=True):
print "Evaluting discriminator pipeline for category", category
evaluated_train = Evaluator("sigmoid_normal3", "ade20k")
evaluated_val = Evaluator("sigmoid_normal3", "ade20k_val")
im_list_train = evaluated_train.get_im_list_by_category(category)
im_list_val = evaluated_val.get_im_list_by_category(category)
print "Train instances:", len(im_list_train)
print "Val instances:", len(im_list_val)
if train:
print "Training..."
self.train(category, im_list_train)
print "Training Done"
else:
print "Skipping training"
print "Running on validation data"
output_fn = "{}-ade20k_val.txt".format(category)
fn_val = self.run(category, im_list_val, output_fn=output_fn)
plot.plot(fn_val, evaluated_val, category, "ade20k_val")
print "Running on training data"
output_fn = "{}-ade20k.txt".format(category)
fn_train = self.run(category, im_list_train, output_fn=output_fn)
plot.plot(fn_train, evaluated_train, category, "ade20k")
plot.vis(fn_val, evaluated_val, category, "ade20k_val",
self.datasource_val)
config = utils.get_config(args.project)
if args.prediction is not None:
config["pspnet_prediction"] = args.prediction
datasource = DataSource(config)
evaluator = Evaluator(args.name, args.project,
datasource) # Evaluation results
vis = ProjectVisualizer(args.project,
datasource,
MAX=args.number,
evaluator=evaluator)
# Image List
im_list = None
if args.im_list:
# Open specific image list
im_list = utils.open_im_list(args.im_list)
elif args.category:
im_list = evaluator.get_im_list_by_category(args.category)
else:
# Open default image list
im_list = utils.open_im_list(config["im_list"])
if args.randomize:
# Shuffle image list
random.seed(3)
random.shuffle(im_list)
im_list = im_list[args.start:]
vis.visualize_images(im_list, category=args.category)
