huangyp/MuseV
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Besides invoke by button, explicitly cal image_length_ratio in gradio

Browse Source
This commit is contained in:
xzqjack 2024-04-11 11:29:08 +08:00
parent 8a81fa6d4e
commit ddba6a4172
3 changed files with 62 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

26
scripts/gradio/app.py
View File

@ -12,6 +12,7 @@ from huggingface_hub import snapshot_download
ProjectDir = os.path.abspath(os.path.join(os.path.dirname(__file__), "../.."))
CheckpointsDir = os.path.join(ProjectDir, "checkpoints")
ignore_video2video = False
max_image_edge = 1280
def download_model():
@ -46,6 +47,9 @@ def hf_online_t2v_inference(
video_len,
img_edge_ratio,
):
img_edge_ratio, _, _ = limit_shape(
image_np, w, h, img_edge_ratio, max_image_edge=max_image_edge
)
if not isinstance(image_np, np.ndarray): # None
raise gr.Error("Need input reference image")
return online_t2v_inference(
@ -66,6 +70,9 @@ def hg_online_v2v_inference(
video_length,
img_edge_ratio,
):
img_edge_ratio, _, _ = limit_shape(
image_np, w, h, img_edge_ratio, max_image_edge=max_image_edge
)
if not isinstance(image_np, np.ndarray): # None
raise gr.Error("Need input reference image")
return online_v2v_inference(
@ -82,11 +89,17 @@ def hg_online_v2v_inference(
)
def limit_shape(image, input_w, input_h, img_edge_ratio, max_image_edge=960):
def limit_shape(image, input_w, input_h, img_edge_ratio, max_image_edge=max_image_edge):
"""limite generation video shape to avoid gpu memory overflow"""
if isinstance(image, np.ndarray) and (input_h == -1 and input_w == -1):
input_h, input_w, _ = image.shape
h, w, _ = image.shape
if input_h == -1 and input_w == -1:
if isinstance(image, np.ndarray):
input_h, input_w, _ = image.shape
elif isinstance(image, PIL.Image.Image):
input_w, input_h = image.size
else:
raise ValueError(
f"image should be in [image, ndarray], but given {type(image)}"
)
if img_edge_ratio == 0:
img_edge_ratio = 1
img_edge_ratio_infact = min(max_image_edge / max(input_h, input_w), img_edge_ratio)
@ -235,8 +248,8 @@ with gr.Blocks(css=css) as demo:
"../../data/images/yongen.jpeg",
],
[
"(masterpiece, best quality, highres:1),(1man, solo:1),(eye blinks:1.8),(head wave:1.3)",
"../../data/images/The-Laughing-Cavalier.jpg",
"(masterpiece, best quality, highres:1), peaceful beautiful sea scene",
"../../data/images/seaside4.jpeg",
],
]
with gr.Row():
@ -345,6 +358,7 @@ with gr.Blocks(css=css) as demo:
fn=hg_online_v2v_inference,
cache_examples=False,
)
img_edge_ratio.change(
fn=limit_shape,
inputs=[image, w, h, img_edge_ratio],

28
scripts/gradio/app_docker_space.py
View File

@ -2,16 +2,19 @@ import os
import time
import pdb
import PIL.Image
import cuid
import gradio as gr
import spaces
import numpy as np
import PIL
from huggingface_hub import snapshot_download
ProjectDir = os.path.abspath(os.path.join(os.path.dirname(__file__), "../.."))
CheckpointsDir = os.path.join(ProjectDir, "checkpoints")
ignore_video2video = True
max_image_edge = 960
def download_model():
@ -46,6 +49,9 @@ def hf_online_t2v_inference(
video_len,
img_edge_ratio,
):
img_edge_ratio, _, _ = limit_shape(
image_np, w, h, img_edge_ratio, max_image_edge=max_image_edge
)
if not isinstance(image_np, np.ndarray): # None
raise gr.Error("Need input reference image")
return online_t2v_inference(
@ -66,6 +72,9 @@ def hg_online_v2v_inference(
video_length,
img_edge_ratio,
):
img_edge_ratio, _, _ = limit_shape(
image_np, w, h, img_edge_ratio, max_image_edge=max_image_edge
)
if not isinstance(image_np, np.ndarray): # None
raise gr.Error("Need input reference image")
return online_v2v_inference(
@ -82,11 +91,17 @@ def hg_online_v2v_inference(
)
def limit_shape(image, input_w, input_h, img_edge_ratio, max_image_edge=960):
def limit_shape(image, input_w, input_h, img_edge_ratio, max_image_edge=max_image_edge):
"""limite generation video shape to avoid gpu memory overflow"""
if isinstance(image, np.ndarray) and (input_h == -1 and input_w == -1):
input_h, input_w, _ = image.shape
h, w, _ = image.shape
if input_h == -1 and input_w == -1:
if isinstance(image, np.ndarray):
input_h, input_w, _ = image.shape
elif isinstance(image, PIL.Image.Image):
input_w, input_h = image.size
else:
raise ValueError(
f"image should be in [image, ndarray], but given {type(image)}"
)
if img_edge_ratio == 0:
img_edge_ratio = 1
img_edge_ratio_infact = min(max_image_edge / max(input_h, input_w), img_edge_ratio)
@ -235,8 +250,8 @@ with gr.Blocks(css=css) as demo:
"../../data/images/yongen.jpeg",
],
[
"(masterpiece, best quality, highres:1),(1man, solo:1),(eye blinks:1.8),(head wave:1.3)",
"../../data/images/The-Laughing-Cavalier.jpg",
"(masterpiece, best quality, highres:1), peaceful beautiful sea scene",
"../../data/images/seaside4.jpeg",
],
]
with gr.Row():
@ -345,6 +360,7 @@ with gr.Blocks(css=css) as demo:
fn=hg_online_v2v_inference,
cache_examples=False,
)
img_edge_ratio.change(
fn=limit_shape,
inputs=[image, w, h, img_edge_ratio],

26
scripts/gradio/app_gradio_space.py
View File

@ -46,6 +46,7 @@ result = subprocess.run(
)
print(result)
ignore_video2video = True
max_image_edge = 960
def download_model():
@ -81,6 +82,9 @@ def hf_online_t2v_inference(
video_len,
img_edge_ratio,
):
img_edge_ratio, _, _ = limit_shape(
image_np, w, h, img_edge_ratio, max_image_edge=max_image_edge
)
if not isinstance(image_np, np.ndarray): # None
raise gr.Error("Need input reference image")
return online_t2v_inference(
@ -101,6 +105,9 @@ def hg_online_v2v_inference(
video_length,
img_edge_ratio,
):
img_edge_ratio, _, _ = limit_shape(
image_np, w, h, img_edge_ratio, max_image_edge=max_image_edge
)
if not isinstance(image_np, np.ndarray): # None
raise gr.Error("Need input reference image")
return online_v2v_inference(
@ -117,11 +124,17 @@ def hg_online_v2v_inference(
)
def limit_shape(image, input_w, input_h, img_edge_ratio, max_image_edge=960):
def limit_shape(image, input_w, input_h, img_edge_ratio, max_image_edge=max_image_edge):
"""limite generation video shape to avoid gpu memory overflow"""
if isinstance(image, np.ndarray) and (input_h == -1 and input_w == -1):
input_h, input_w, _ = image.shape
h, w, _ = image.shape
if input_h == -1 and input_w == -1:
if isinstance(image, np.ndarray):
input_h, input_w, _ = image.shape
elif isinstance(image, PIL.Image.Image):
input_w, input_h = image.size
else:
raise ValueError(
f"image should be in [image, ndarray], but given {type(image)}"
)
if img_edge_ratio == 0:
img_edge_ratio = 1
img_edge_ratio_infact = min(max_image_edge / max(input_h, input_w), img_edge_ratio)
@ -270,8 +283,8 @@ with gr.Blocks(css=css) as demo:
"../../data/images/yongen.jpeg",
],
[
"(masterpiece, best quality, highres:1),(1man, solo:1),(eye blinks:1.8),(head wave:1.3)",
"../../data/images/The-Laughing-Cavalier.jpg",
"(masterpiece, best quality, highres:1), peaceful beautiful sea scene",
"../../data/images/seaside4.jpeg",
],
]
with gr.Row():
@ -380,6 +393,7 @@ with gr.Blocks(css=css) as demo:
fn=hg_online_v2v_inference,
cache_examples=False,
)
img_edge_ratio.change(
fn=limit_shape,
inputs=[image, w, h, img_edge_ratio],