SignLanguage / app.py
ZiyuG's picture
Update app.py
155fa4f verified
import gradio as gr
from evaluate import eval, install
from datetime import datetime
import os, shutil, spaces
import warnings, tempfile
from audio import gen_audio
warnings.filterwarnings("ignore")
warnings.simplefilter("ignore")
import time, cv2, os
videos = {
# "教學視頻1": "teach/00.mp4",
# "教學視頻2": "teach/01.mp4",
# "教學視頻3": "teach/02.mp4",
# "教學視頻4": "teach/03.mp4",
"(病症)高峰期": "teach/(病症)高峰期.mp4",
"公園": "teach/公園.mp4",
"健康": "teach/健康.mp4",
"刷牙": "teach/刷牙.mp4",
"圖書館": "teach/圖書館.mp4",
"下午": "teach/下午.mp4",
"中風": "teach/中風.mp4",
"中暑": "teach/中暑.mp4",
"主題": "teach/主題.mp4",
"住院證明": "teach/住院證明.mp4"
}
jsons = {
# "教學視頻1": "teach/00.json",
# "教學視頻2": "teach/01.json",
# "教學視頻3": "teach/02.json",
# "教學視頻4": "teach/03.json",
"(病症)高峰期": "teach/(病症)高峰期.json",
"公園": "teach/公園.json",
"健康": "teach/健康.json",
"刷牙": "teach/刷牙.json",
"圖書館": "teach/圖書館.json",
"下午": "teach/下午.json",
"中風": "teach/中風.json",
"中暑": "teach/中暑.json",
"主題": "teach/主題.json",
"住院證明": "teach/住院證明.json"
}
def get_video_url(option):
# print(videos[option])
return videos[option]
def flip_video_horizontally(video_path):
if not os.path.exists(video_path):
print(f"Error: {video_path} does not exist.")
return
cap = cv2.VideoCapture(video_path)
if not cap.isOpened():
print(f"Error: cannot open {video_path}.")
return
frame_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
frame_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
fps = int(cap.get(cv2.CAP_PROP_FPS))
fourcc = cv2.VideoWriter_fourcc(*'XVID') #int(cap.get(cv2.CAP_PROP_FOURCC))
temp_output_path = video_path + "_temp.mp4"
out = cv2.VideoWriter(temp_output_path, fourcc, fps, (frame_width, frame_height))
while True:
ret, frame = cap.read()
if not ret:
break
flipped_frame = cv2.flip(frame, 1)
out.write(flipped_frame)
cap.release()
out.release()
print(video_path, os.path.exists(video_path))
print(temp_output_path, os.path.exists(temp_output_path))
os.replace(temp_output_path, video_path)
print(f"Finished flipping video:{video_path}")
@spaces.GPU()
def evaluate_sign_language(user_video, standard_video_option):
if user_video.split('/')[-1].startswith("sample."):
flip_video_horizontally(user_video)
tmp = tempfile.TemporaryDirectory()
tmpdir = tmp.name
new_path = tmpdir + "/user.mp4"
shutil.copy(user_video, new_path)
shutil.copy(videos[standard_video_option], tmpdir + "/standard.mp4")
shutil.copy(jsons[standard_video_option], tmpdir + "/standard.json")
print("Copy User Video to:", new_path)
print("Copy Standard Video to:", tmpdir + "/standard.mp4")
print("Copy Standard JSON to:", tmpdir + "/standard.json")
test, standard = user_video, videos[standard_video_option]
score, final_merged_intervals, comments = eval(test, standard, tmpdir)
if score < 95:
score = 0
elif score < 98:
score = 50 * (score - 95) / 3
elif 98 <= score <= 100:
score = 50 * (score - 98) / 2 + 50
current_date = datetime.now().strftime("%Y年%m月%d日")
# 初始化反馈报告
qualification = "不合格" if score < 60 else "合格"
advice = f"""<div style='font-family: "Heiti SC"; text-align: center; font-size: 24px;'><b>\n手語表現反饋報告</b></div>
<br><br><br><b>日期:</b> {current_date}<br><br><b>改進建議(供參考):</b><br>"""
part_translation = {
'Right Hand': '右手細節、',
'Left Hand': '左手細節、',
'Right Arm': '右臂、',
'Left Arm': '左臂、'
}
# 为每个问题区间生成具体观察结果
f = 0
subtitles = []
for interval, errors in final_merged_intervals:
advice += f"<b> - 時間區間:{interval[0]}-{interval[1]}秒</b><br>"
parts = ""
for error in errors:
part = part_translation[error]
parts += part
advice += f" 觀察結果:您的{parts[:-1]}動作與標準手語手勢不太一致。<br>"
f = 1
subtitles.append([[interval[0], interval[1]], parts[:-1]])
if not f:
advice = advice[:-4]
advice += "無"
advice += "<br>建議:請參考標準視頻,模仿正確的動作,保持速度一致。<br><br>"
advice += f"<b>整體结果:</b> {qualification}<br>"
# advice += f"<b>整體结果:</b> 您本次的總評分為 {score:.2f}/100分。<br>"
ret_video = gen_audio(subtitles, tmpdir)
return advice, ret_video, tmp
def cleanup_temp_dir(tmp_dir):
if tmp_dir:
time.sleep(0.5)
print(f"Cleaning up temporary directory: {tmp_dir.name}")
tmp_dir.cleanup()
return "临时目录已清理"
return "没有临时目录需要清理"
install()
font = ["Heiti SC", "FangSong"]
title = """<h1 style="text-align: center;">
<div style="width: 1.4em; height: 1.4em; display: inline-block;"><img src="https://github.com/ZiyuGuo99/ZiyuGuo99.github.io/blob/main/assets/img/sl.png?raw=true" style='width: 100%; height: 100%; object-fit: contain;' /></div>
<span style="font-family: """ + font[0] + """; font-variant: small-caps; font-weight: bold;">手語教學與評估系統</span>
</h1>
"""
with gr.Blocks(
css="""
.contain { display: flex; flex-direction: column; }
.gradio-container { height: 200vh !important;
overflow-y: auto; /* 添加垂直滚动条 */
}
#col_container { height: 100%; }
pre {
white-space: pre-wrap; /* Since CSS 2.1 */
white-space: -moz-pre-wrap; /* Mozilla, since 1999 */
white-space: -pre-wrap; /* Opera 4-6 */
white-space: -o-pre-wrap; /* Opera 7 */
word-wrap: break-word; /* Internet Explorer 5.5+ */
}""",
js="""
function refresh() {
const url = new URL(window.location);
if (url.searchParams.get('__theme') !== 'light') {
url.searchParams.set('__theme', 'light');
window.location.href = url.href;
}
}""",
title="手語教學與評估系統",
theme=gr.themes.Soft(),
) as app:
gr.HTML(title)
with gr.Row():
with gr.Column():
video_selector = gr.Dropdown(list(videos.keys()), label="請選擇教學視頻")
video_player = gr.Video(label="請觀看教學視頻演示")
with gr.Column():
upload_video = gr.Video(label="請錄製/上傳您的視頻,點擊\'開始評估\'")
evaluate_button = gr.Button("開始評估")
with gr.Row():
with gr.Column():
advice_output = gr.HTML(label="反饋報告")
with gr.Column():
compare_video_player = gr.Video(label="評估結果")
video_selector.change(get_video_url, inputs=video_selector, outputs=video_player)
tmp_dir_state = gr.State()
evaluate_button.click(
evaluate_sign_language,
inputs=[upload_video, video_selector],
outputs=[advice_output, compare_video_player, tmp_dir_state],
)
compare_video_player.change(
fn=cleanup_temp_dir,
inputs=tmp_dir_state,
outputs=None
)
app.launch(share=True)