ArcaneSVK2 / app.py
Podtekatel's picture
Added new model
de7b3c0
raw
history blame
2.92 kB
import logging
import os
import gradio as gr
import numpy as np
from PIL import Image
from huggingface_hub import hf_hub_url, cached_download
from inference.face_detector import StatRetinaFaceDetector
from inference.model_pipeline import VSNetModelPipeline
from inference.onnx_model import ONNXModel
logging.basicConfig(
format='%(asctime)s %(levelname)-8s %(message)s',
level=logging.INFO,
datefmt='%Y-%m-%d %H:%M:%S')
MODEL_IMG_SIZE = 256
def load_model():
REPO_ID = "Podtekatel/ARCNEGAN"
FILENAME_OLD = "arcane_exp_203_ep_399.onnx"
FILENAME_NEW = "arcane_exp_206_ep_138.onnx"
global model_old
global model_new
global pipeline_old
global pipeline_new
# Old model
model_path = cached_download(
hf_hub_url(REPO_ID, FILENAME_OLD), use_auth_token=os.getenv('HF_TOKEN')
)
model_old = ONNXModel(model_path)
pipeline_old = VSNetModelPipeline(model_old, StatRetinaFaceDetector(MODEL_IMG_SIZE), background_resize=1024, no_detected_resize=1024)
# New model
model_path = cached_download(
hf_hub_url(REPO_ID, FILENAME_NEW), use_auth_token=os.getenv('HF_TOKEN')
)
model_new = ONNXModel(model_path)
pipeline_new = VSNetModelPipeline(model_new, StatRetinaFaceDetector(MODEL_IMG_SIZE), background_resize=1024,
no_detected_resize=1024)
return model_old, model_new
load_model()
def inference(img, ver):
img = np.array(img)
if ver == 'version 2':
out_img = pipeline_new(img)
else:
out_img = pipeline_old(img)
out_img = Image.fromarray(out_img)
return out_img
title = "ARCNStyleTransfer"
description = "Gradio Demo for Arcane Season 1 style transfer. To use it, simply upload your image, or click one of the examples to load them."
article = "This is one of my successful experiments on style transfer. I've built my own pipeline, generator model and private dataset to train this model<br>" \
"" \
"" \
"" \
"Model pipeline which used in project is improved CartoonGAN.<br>" \
"This model was trained on RTX 2080 Ti 1.5 days with batch size 7.<br>" \
"Model weights 64 MB in ONNX fp32 format, infers 25 ms on GPU and 150 ms on CPU at 256x256 resolution.<br>" \
"If you want to use this app or integrate this model into yours, please contact me at email '[email protected]'."
imgs_folder = 'demo'
examples = [[os.path.join(imgs_folder, img_filename), version] for img_filename in sorted(os.listdir(imgs_folder)) for version in ['version 2']]
demo = gr.Interface(
fn=inference,
inputs=[gr.inputs.Image(type="pil"), gr.inputs.Radio(['version 1', 'version 2'], type="value", default='version 2', label='version')],
outputs=gr.outputs.Image(type="pil"),
title=title,
description=description,
article=article,
examples=examples)
demo.launch()