Skip to content

Commit

Permalink
Add 2 WebUI Sample apps
Browse files Browse the repository at this point in the history
  • Loading branch information
@quic-zhanweiw
quic-zhanweiw committed Feb 6, 2025
1 parent 5bc8341 commit 7b186f7
Show file tree
Hide file tree
Showing 6 changed files with 635 additions and 52 deletions.
67 changes: 35 additions & 32 deletions samples/python/real_esrgan_x4plus/real_esrgan_x4plus.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -14,9 +14,15 @@
import torchvision.transforms as transforms
from PIL import Image
from PIL.Image import fromarray as ImageFromArray

from utils.image_processing import (
preprocess_PIL_image,
torch_tensor_to_PIL_image,
pil_resize_pad,
pil_undo_resize_pad
)
from qai_appbuilder import (QNNContext, Runtime, LogLevel, ProfilingLevel, PerfProfile, QNNConfig)


####################################################################

MODEL_ID = "mnz1l2exq"
Expand All @@ -27,6 +33,10 @@
####################################################################

execution_ws = os.getcwd()

if not "python" in execution_ws:
execution_ws = execution_ws + "\\..\\" + "python"

qnn_dir = execution_ws + "\\qai_libs"

if not MODEL_NAME in execution_ws:
Expand All @@ -35,27 +45,12 @@
model_dir = execution_ws + "\\models"
madel_path = model_dir + "\\" + MODEL_NAME + ".bin"


####################################################################

image_buffer = None
realesrgan = None

def preprocess_PIL_image(image: Image) -> torch.Tensor:
"""Convert a PIL image into a pyTorch tensor with range [0, 1] and shape NCHW."""
transform = transforms.Compose([transforms.Resize(IMAGE_SIZE), # bgr image
transforms.CenterCrop(IMAGE_SIZE),
transforms.PILToTensor()])
img: torch.Tensor = transform(image) # type: ignore
img = img.float() / 255.0 # int 0 - 255 to float 0.0 - 1.0
return img

def torch_tensor_to_PIL_image(data: torch.Tensor) -> Image:
"""
Convert a Torch tensor (dtype float32) with range [0, 1] and shape CHW into PIL image CHW
"""
out = torch.clip(data, min=0.0, max=1.0)
np_out = (out.detach().numpy() * 255).astype(np.uint8)
return ImageFromArray(np_out)

# RealESRGan class which inherited from the class QNNContext.
class RealESRGan(QNNContext):
Expand Down Expand Up @@ -85,13 +80,15 @@ def Init():
# Instance for RealESRGan objects.
realesrgan = RealESRGan("realesrgan", madel_path)

def Inference(input_image_path, output_image_path):
def Inference(input_image_path, output_image_path, show_image = True):
global image_buffer

# Read and preprocess the image.
image = Image.open(input_image_path)
image = preprocess_PIL_image(image).numpy()
image = np.transpose(image, (1, 2, 0)) # CHW -> HWC
orig_image = Image.open(input_image_path)
image, scale, padding = pil_resize_pad(orig_image, (IMAGE_SIZE, IMAGE_SIZE))

image = np.array(image)
image = np.clip(image, 0, 255) / 255.0 # normalization

# Burst the HTP.
PerfProfile.SetPerfProfileGlobal(PerfProfile.BURST)
Expand All @@ -102,25 +99,31 @@ def Inference(input_image_path, output_image_path):
# Reset the HTP.
PerfProfile.RelPerfProfileGlobal()

# show & save the result
output_image = torch.from_numpy(output_image)
output_image = output_image.reshape(IMAGE_SIZE * 4, IMAGE_SIZE * 4, 3)
output_image = torch.unsqueeze(output_image, 0)
output_image = [torch_tensor_to_PIL_image(img) for img in output_image]
image_buffer = output_image[0]

output_image = np.clip(output_image, 0.0, 1.0)
output_image = (output_image * 255).astype(np.uint8) # un-normalization
output_image = ImageFromArray(output_image)

image_size = (orig_image.size[0] * 4, orig_image.size[1] * 4)
image_padding = (padding[0] * 4, padding[1] * 4)
image_buffer = pil_undo_resize_pad(output_image, image_size, scale, image_padding)

# show & save the result
image_buffer.save(output_image_path)
image_buffer.show()

if show_image:
image_buffer.show()

def Release():
global realesrgan

# Release the resources.
del(realesrgan)

if __name__ == '__main__':
Init()

Init()

Inference(execution_ws + "\\input.jpg", execution_ws + "\\output.jpg")

Release()
Inference(execution_ws + "\\input.jpg", execution_ws + "\\output.jpg")

Release()
30 changes: 20 additions & 10 deletions samples/python/stable_diffusion_v1_5/stable_diffusion_v1_5.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -45,6 +45,10 @@
####################################################################

execution_ws = os.getcwd()

if not "python" in execution_ws:
execution_ws = execution_ws + "\\..\\" + "python"

qnn_dir = execution_ws + "\\qai_libs"

if not MODEL_NAME in execution_ws:
Expand Down Expand Up @@ -117,6 +121,10 @@ def model_initialize():

result = True

SetQNNConfig()

model_download()

# model names
model_text_encoder = "text_encoder"
model_unet = "model_unet"
Expand Down Expand Up @@ -166,13 +174,15 @@ def setup_parameters(prompt, un_prompt, seed, step, text_guidance):

user_prompt = prompt
uncond_prompt = un_prompt
user_seed = seed
user_seed = np.int64(seed)
user_step = step
user_text_guidance = text_guidance

if user_seed == -1:
user_seed = np.random.randint(low=0, high=9999999999, size=None, dtype=np.int64)

assert isinstance(user_seed, np.int64) == True, "user_seed should be of type int64"
assert isinstance(user_step, int) == True, "user_step should be of type int"
assert user_step == 20 or user_step == 30 or user_step == 50, "user_step should be either 20, 30 or 50"
assert isinstance(user_text_guidance, float) == True, "user_text_guidance should be of type float"
assert user_text_guidance >= 5.0 and user_text_guidance <= 15.0, "user_text_guidance should be a float from [5.0, 15.0]"

Expand Down Expand Up @@ -209,7 +219,7 @@ def get_time_embedding(timestep, time_embeddings):
return emb

# Execute the Stable Diffusion pipeline
def model_execute(callback):
def model_execute(callback, image_path, show_image = True):
PerfProfile.SetPerfProfileGlobal(PerfProfile.BURST)

scheduler.set_timesteps(user_step) # Setting up user provided time steps for Scheduler
Expand Down Expand Up @@ -259,7 +269,7 @@ def model_execute(callback):
callback(None)
else:
image_size = 512
image_path = execution_ws + "\\images"

if not os.path.exists(image_path):
os.makedirs(image_path, exist_ok=True)
image_path = image_path + "\\%s_%s_%s.jpg"%(formatted_time, str(user_seed), str(image_size))
Expand All @@ -268,12 +278,16 @@ def model_execute(callback):
output_image = output_image.reshape(image_size, image_size, -1)
output_image = Image.fromarray(output_image, mode="RGB")
output_image.save(image_path)
output_image.show()

if show_image:
output_image.show()

callback(image_path)

PerfProfile.RelPerfProfileGlobal()

return image_path

# Release all the models.
def model_destroy():
global text_encoder
Expand Down Expand Up @@ -332,10 +346,6 @@ def model_download():
parser.add_argument("--prompt", default=DEFAULT_PROMPT, type=str)
args = parser.parse_args()

SetQNNConfig()

model_download()

model_initialize()

time_start = time.time()
Expand All @@ -347,7 +357,7 @@ def model_download():
user_text_guidance = 7.5

setup_parameters(user_prompt, uncond_prompt, user_seed, user_step, user_text_guidance)
model_execute(modelExecuteCallback)
model_execute(modelExecuteCallback, execution_ws + "\\images")

time_end = time.time()
print("time consumes for inference {}(s)".format(str(time_end - time_start)))
Expand Down
30 changes: 20 additions & 10 deletions samples/python/stable_diffusion_v2_1/stable_diffusion_v2_1.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -45,6 +45,10 @@
####################################################################

execution_ws = os.getcwd()

if not "python" in execution_ws:
execution_ws = execution_ws + "\\..\\" + "python"

qnn_dir = execution_ws + "\\qai_libs"

if not MODEL_NAME in execution_ws:
Expand Down Expand Up @@ -117,6 +121,10 @@ def model_initialize():

result = True

SetQNNConfig()

model_download()

# model names
model_text_encoder = "text_encoder"
model_unet = "model_unet"
Expand Down Expand Up @@ -166,13 +174,15 @@ def setup_parameters(prompt, un_prompt, seed, step, text_guidance):

user_prompt = prompt
uncond_prompt = un_prompt
user_seed = seed
user_seed = np.int64(seed)
user_step = step
user_text_guidance = text_guidance

if user_seed == -1:
user_seed = np.random.randint(low=0, high=9999999999, size=None, dtype=np.int64)

assert isinstance(user_seed, np.int64) == True, "user_seed should be of type int64"
assert isinstance(user_step, int) == True, "user_step should be of type int"
assert user_step == 20 or user_step == 30 or user_step == 50, "user_step should be either 20, 30 or 50"
assert isinstance(user_text_guidance, float) == True, "user_text_guidance should be of type float"
assert user_text_guidance >= 5.0 and user_text_guidance <= 15.0, "user_text_guidance should be a float from [5.0, 15.0]"

Expand Down Expand Up @@ -201,7 +211,7 @@ def get_time_embedding(timestep, time_embeddings):
return emb

# Execute the Stable Diffusion pipeline
def model_execute(callback):
def model_execute(callback, image_path, show_image = True):
PerfProfile.SetPerfProfileGlobal(PerfProfile.BURST)

scheduler.set_timesteps(user_step) # Setting up user provided time steps for Scheduler
Expand Down Expand Up @@ -250,7 +260,7 @@ def model_execute(callback):
callback(None)
else:
image_size = 512
image_path = execution_ws + "\\images"

if not os.path.exists(image_path):
os.makedirs(image_path, exist_ok=True)
image_path = image_path + "\\%s_%s_%s.jpg"%(formatted_time, str(user_seed), str(image_size))
Expand All @@ -259,12 +269,16 @@ def model_execute(callback):
output_image = output_image.reshape(image_size, image_size, -1)
output_image = Image.fromarray(output_image, mode="RGB")
output_image.save(image_path)
output_image.show()

if show_image:
output_image.show()

callback(image_path)

PerfProfile.RelPerfProfileGlobal()

return image_path

# Release all the models.
def model_destroy():
global text_encoder
Expand Down Expand Up @@ -323,10 +337,6 @@ def model_download():
parser.add_argument("--prompt", default=DEFAULT_PROMPT, type=str)
args = parser.parse_args()

SetQNNConfig()

model_download()

model_initialize()

time_start = time.time()
Expand All @@ -338,7 +348,7 @@ def model_download():
user_text_guidance = 7.5

setup_parameters(user_prompt, uncond_prompt, user_seed, user_step, user_text_guidance)
model_execute(modelExecuteCallback)
model_execute(modelExecuteCallback, execution_ws + "\\images")

time_end = time.time()
print("time consumes for inference {}(s)".format(str(time_end - time_start)))
Expand Down
Loading

0 comments on commit 7b186f7

Please sign in to comment.