def run(self):
while True:
# Perform style transfer for payload
payload = self.queue.get()
# Unpack and setup style transer for payload
content_data, style_data, tag, settings = api.unpack_payload(
payload)
content_image = convert_image(content_data)
style_image = convert_image(style_data)
# Perform style transfer
if self.verbose: print("[StyleWorker]: processing payload: ", tag)
n_epochs = 100
if api.SETTING_NUMBER_EPOCHS_KEY in settings:
n_epochs = settings[api.SETTING_NUMBER_EPOCHS_KEY]
pastiche_image = styleopt.transfer_style(content_image,
style_image,
n_epochs=n_epochs,
settings=settings,
verbose=self.verbose)
# Save results of style transfer
if self.verbose: print("[StyleWorker]: completed payload: ", tag)
if not os.path.exists("static/pastiche"):
os.mkdir("static/pastiche")
pastiche_image.save("static/pastiche/{}.jpg".format(tag))
def get_logo_id_new(logo_url, download_crawler, source, sourceId, catename):
mongo = db.connect_mongo()
# imgfs = gridfs.GridFS(mongo.gridfs)
name = None
height = None
width = None
if logo_url is not None and len(logo_url.strip()) > 0:
logger.info("Download logo: %s", logo_url)
# (image_value, width, height) = download_crawler.get_image_size(logo_url)
(image_file, width,
height) = download_crawler.get_image_size_new(logo_url)
if image_file is not None:
# logo_id = imgfs.put(image_value, content_type='jpeg', filename='%s_%s_%s.jpg' % (catename, source, sourceId))
# out = imgfs.get(ObjectId(logo_id))
name = util.get_uuid()
logger.info("%s->%s|%s", logo_url, name, image_file)
if source in [13835, 13836, 13613]:
# img, width, height = util.convert_image(out, out.name, size=1024)
img, width, height = util.convert_image(image_file,
name,
size=1024)
elif source in [13613, 13803]:
img, width, height = util.convert_image(image_file,
name,
size=width)
else:
# img, width, height = util.convert_image(out, out.name)
img, width, height = util.convert_image(image_file, name)
headers = {"Content-Type": "image/jpeg"}
# oss2put.put(str(logo_id), img, headers=headers)
oss2put.put(name, img, headers=headers)
# mongo.close()
return (name, width, height)
def get_page_image(workflow, seq_num, img_type, plugname):
""" Return image for requested page. """
if img_type not in ('raw', 'processed'):
raise ApiException("Image type must be one of 'raw' or 'processed', "
"not '{0}'".format(img_type), 400)
# Scale image if requested
width = request.args.get('width', None)
img_format = request.args.get('format', None)
page = get_next(p for p in workflow.pages if p.sequence_num == seq_num)
if not page:
raise ApiException("Could not find page with sequence number {0}"
.format(seq_num), 404)
if img_type == 'raw':
fpath = page.raw_image
elif plugname is None:
fpath = page.get_latest_processed(image_only=True)
else:
fpath = page.processed_images[plugname]
if width and fpath.suffix.lower() in ('.jpg', '.jpeg', '.tif', '.tiff',
'.png'):
return scale_image(fpath, width=int(width))
elif fpath.suffix.lower() in ('.tif', '.tiff') and img_format:
img_format = 'png' if img_format == 'browser' else img_format
return convert_image(fpath, img_format)
else:
return send_file(unicode(fpath))
def get_page_image(workflow, seq_num, img_type, plugname):
""" Return image for requested page. """
if img_type not in ('raw', 'processed'):
raise ApiException("Image type must be one of 'raw' or 'processed', "
"not '{0}'".format(img_type), 400)
# Scale image if requested
width = request.args.get('width', None)
img_format = request.args.get('format', None)
page = get_next(p for p in workflow.pages if p.sequence_num == seq_num)
if not page:
raise ApiException("Could not find page with sequence number {0}"
.format(seq_num), 404)
if img_type == 'raw':
fpath = page.raw_image
elif plugname is None:
fpath = page.get_latest_processed(image_only=True)
else:
fpath = page.processed_images[plugname]
if width and fpath.suffix.lower() in ('.jpg', '.jpeg', '.tif', '.tiff',
'.png'):
return scale_image(fpath, width=int(width))
elif fpath.suffix.lower() in ('.tif', '.tiff') and img_format:
img_format = 'png' if img_format == 'browser' else img_format
return convert_image(fpath, img_format)
else:
return send_file(unicode(fpath))
def save_oss2_image(grid_id, size=None):
if grid_id is None or grid_id.strip() == "":
return
item = mongo.temp.gridid.find_one({"gridid": grid_id})
if item is not None:
return
out = grid.get(ObjectId(grid_id))
logger.info("%s -> %s", grid_id, out.name)
if size is None:
img, xsize, ysize = util.convert_image(out, out.name)
else:
img, xsize, ysize = util.convert_image(out, out.name, size=size)
headers = {"Content-Type": "image/jpeg"}
oss2.put(grid_id, img, headers=headers)
mongo.temp.gridid.insert({"gridid": grid_id})
def get_page_image(fpath, page, workflow, number, img_type, plugname):
""" Get image for requested page.
:param workflow: UUID or slug for a workflow
:type workflow: str
:param number: Capture number of requested page
:type number: int
:param img_type: Type of image
:type img_type: str, one of `raw` or `processed`
:param plugname: Only applicable if `img_type` is `processed`,
selects the desired processed file by its key in the
:py:attr:`spreads.workflow.Workflow.processed_images`
dictionary.
:type plugname: str
:queryparam width: Optionally scale down image to the desired width
:type width: int
:queryparam format: Optionally convert image to desired format.
If `browser` is specified, non-JPG or PNG images will
be converted to PNG.
:type format: str, either `browser` or a format string recognized
by Pillow: http://pillow.readthedocs.org/en/latest/\\
handbook/image-file-formats.html
:resheader Content-Type: Depends on value of `format`, by default
the mime-type of the original image.
"""
width = request.args.get('width', None)
img_format = request.args.get('format', None)
# FIXME: This clearly sucks, rework convert_image and scale_image to allow
# for it.
if width is not None and img_format is not None:
raise ApiException("Can not scale and convert at the same time.", 400)
transformable = fpath.suffix.lower() in ('.jpg', '.jpeg', '.tif', '.tiff',
'.png')
if (width is not None or img_format is not None) and not transformable:
raise ApiException("Can only scale/convert JPG, TIF or PNG files.",
400)
if width:
# Scale image if requested
return scale_image(fpath, width=int(width))
elif img_format:
# Convert to target format
if fpath.suffix.lower() not in ('.tif', '.tiff', '.jpg', '.jpeg'):
img_format = 'png' if img_format == 'browser' else img_format
return convert_image(fpath, img_format)
else:
# Send unmodified if no scaling/converting is requested
return send_file(unicode(fpath))
def get_page_image(workflow, number, img_type, plugname):
""" Return image for requested page. """
# Scale image if requested
width = request.args.get('width', None)
img_format = request.args.get('format', None)
page = find_page(workflow, number)
check_page_params(page, img_type, plugname)
if img_type == 'raw':
fpath = page.raw_image
elif plugname is None:
fpath = page.get_latest_processed(image_only=True)
else:
fpath = page.processed_images[plugname]
if width and fpath.suffix.lower() in ('.jpg', '.jpeg', '.tif', '.tiff',
'.png'):
return scale_image(fpath, width=int(width))
elif fpath.suffix.lower() in ('.tif', '.tiff') and img_format:
img_format = 'png' if img_format == 'browser' else img_format
return convert_image(fpath, img_format)
else:
return send_file(unicode(fpath))
def test():
if opt.specific_observation_idcs is not None:
specific_observation_idcs = list(
map(int, opt.specific_observation_idcs.split(',')))
else:
specific_observation_idcs = None
dataset = dataio.SceneClassDataset(
root_dir=opt.data_root,
max_num_instances=opt.max_num_instances,
specific_observation_idcs=specific_observation_idcs,
max_observations_per_instance=-1,
samples_per_instance=1,
img_sidelength=opt.img_sidelength)
dataset = DataLoader(dataset,
collate_fn=dataset.collate_fn,
batch_size=1,
shuffle=False,
drop_last=False)
model = SRNsModel(num_instances=opt.num_instances,
latent_dim=opt.embedding_size,
has_params=opt.has_params,
fit_single_srn=opt.fit_single_srn,
use_unet_renderer=opt.use_unet_renderer,
tracing_steps=opt.tracing_steps)
assert (opt.checkpoint_path is not None), "Have to pass checkpoint!"
print("Loading model from %s" % opt.checkpoint_path)
util.custom_load(model,
path=opt.checkpoint_path,
discriminator=None,
overwrite_embeddings=False)
model.eval()
model.cuda()
# directory structure: month_day/
renderings_dir = os.path.join(opt.logging_root, 'renderings')
gt_comparison_dir = os.path.join(opt.logging_root, 'gt_comparisons')
util.cond_mkdir(opt.logging_root)
util.cond_mkdir(gt_comparison_dir)
util.cond_mkdir(renderings_dir)
# Save command-line parameters to log directory.
with open(os.path.join(opt.logging_root, "params.txt"), "w") as out_file:
out_file.write('\n'.join(
["%s: %s" % (key, value) for key, value in vars(opt).items()]))
print('Beginning evaluation...')
with torch.no_grad():
instance_idx = 0
idx = 0
psnrs, ssims = list(), list()
for model_input, ground_truth in dataset:
model_outputs = model(model_input)
psnr, ssim = model.get_psnr(model_outputs, ground_truth)
psnrs.extend(psnr)
ssims.extend(ssim)
instance_idcs = model_input['instance_idx']
print("Object instance %d. Running mean PSNR %0.6f SSIM %0.6f" %
(instance_idcs[-1], np.mean(psnrs), np.mean(ssims)))
if instance_idx < opt.save_out_first_n:
output_imgs = model.get_output_img(model_outputs).cpu().numpy()
comparisons = model.get_comparisons(model_input, model_outputs,
ground_truth)
for i in range(len(output_imgs)):
prev_instance_idx = instance_idx
instance_idx = instance_idcs[i]
if prev_instance_idx != instance_idx:
idx = 0
img_only_path = os.path.join(renderings_dir,
"%06d" % instance_idx)
comp_path = os.path.join(gt_comparison_dir,
"%06d" % instance_idx)
util.cond_mkdir(img_only_path)
util.cond_mkdir(comp_path)
pred = util.convert_image(output_imgs[i].squeeze())
comp = util.convert_image(comparisons[i].squeeze())
util.write_img(
pred, os.path.join(img_only_path, "%06d.png" % idx))
util.write_img(comp,
os.path.join(comp_path, "%06d.png" % idx))
idx += 1
with open(os.path.join(opt.logging_root, "results.txt"), "w") as out_file:
out_file.write("%0.6f, %0.6f" % (np.mean(psnrs), np.mean(ssims)))
print("Final mean PSNR %0.6f SSIM %0.6f" %
(np.mean(psnrs), np.mean(ssims)))
if options["verbose"]:
print("Sending style transfer request to server... ", end="")
r = requests.post(api.SERVER_URL + "/api/style", json=payload)
# Wait for success response from server
if options["verbose"]: print("Waiting for response from server ", end="")
has_pastiche = False
while not has_pastiche:
r = requests.get(api.SERVER_URL + "/api/pastiche/" + tag_id)
if r.status_code == api.STATUS_OK:
print(" Done!")
has_pastiche = True
elif r.status_code == api.STATUS_NOT_READY:
print("#", end="", flush=True)
time.sleep(1)
else:
raise ValueError("FATAL: something went wrong")
# Read pastiche image from server
if options["verbose"]: print("Loading pastiche from server... ", end="")
r = requests.get(api.SERVER_URL + "/api/pastiche/" + tag_id)
if r.status_code == api.STATUS_OK:
pastiche_image = convert_image(r.content)
else:
raise ValueError("FATAL: something went wrong")
# Show pastiche image
plt.imshow(np.asarray(pastiche_image))
plt.show()
if CAPTURE_MODE:
# capture an image every second, and save as a training sample
if frame_count % FPS == 0:
ROI = depth[top_left[1] + 2:bottom_right[1],
top_left[0] + 2:bottom_right[0]]
#cv2.imshow('img', ROI)
util.capture_image(ROI, img_count, IMG_SAVE_PATH)
img_count += 1
else:
# capture every 0.5 second
if frame_count % FPS == 0 or frame_count % FPS == 0.5 * FPS:
ROI = depth[top_left[1] + 2:bottom_right[1],
top_left[0] + 2:bottom_right[0]]
img = util.resize_and_smooth(ROI)
img = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB)
img = util.convert_image(img)
prediction = model.predict(img)
arg = np.argmax(prediction)
number = mapping[arg]
if number != last_prediction:
last_prediction = number
label.config(text=str(number))
print("Predicted number: {}".format(str(number)))
cv2.imshow("depth image", depth)
if enable_rgb:
cv2.imshow("color",
cv2.resize(color.asarray(), (int(1920 / 3), int(1080 / 3))))
listener.release(frames)
key = cv2.waitKey(delay=1)
if key == ord('q'):
