Gradiodemo.py extension: selectable outputs
Opened this issue · 0 comments
bananaman1983 commented
Added radioboxes to implement the support for the other output formats available in the demo.py
Results are stored in ./examples/results folder with 5 digit randomized serial tags. Chances are low but files may get overwritten so don't pile up on the outputs.
A little like a note to my self
Implementation to windows os is possible if bit tricky with all the lost dependencies and stuff.
A little more tweak on the code is necessary beyond what is aforementioned at the readme.md
-
MSVC alone wont be enough for the command line build. check for cl.exe and if it's included in the path env. variables
-
long is deprecated in the recent numpy releases. np.long should be converted to np.longlong in the ./bfm/bfm.py
-
due to the library collision, additional line was necessary at the header for the gradiodemo.py to work
os.environ['KMP_DUPLICATE_LIB_OK'] = 'True'
- there is no 'input'/'output' attribute in gradio
gradiodemo.py
# before import, make sure FaceBoxes and Sim3DR are built successfully, e.g.,
import sys
from subprocess import call
import os
import torch
os.environ['KMP_DUPLICATE_LIB_OK']='True'
import random
import string
torch.hub.download_url_to_file('https://upload.wikimedia.org/wikipedia/commons/thumb/6/6e/Solvay_conference_1927.jpg/1400px-Solvay_conference_1927.jpg', 'solvay.jpg')
def run_cmd(command):
try:
print(command)
call(command, shell=True)
except Exception as e:
print(f"Errorrrrr: {e}!")
print(os.getcwd())
#os.chdir("/FaceBoxes/utils")
print(os.getcwd())
#run_cmd("python build.py build_ext --inplace")
#os.chdir("/Sim3DR")
print(os.getcwd())
#run_cmd("python setup.py build_ext --inplace")
print(os.getcwd())
#os.chdir("/utils/asset")
print(os.getcwd())
#run_cmd("gcc -shared -Wall -O3 render.c -o render.so -fPIC")
#os.chdir("/app")
print(os.getcwd())
import cv2
import yaml
from FaceBoxes import FaceBoxes
from TDDFA import TDDFA
from utils.render import render
from utils.depth import depth
from utils.pncc import pncc
from utils.uv import uv_tex
from utils.pose import viz_pose
from utils.serialization import ser_to_ply, ser_to_obj
from utils.functions import draw_landmarks, get_suffix
from utils.tddfa_util import str2bool
import matplotlib.pyplot as plt
from skimage import io
import gradio as gr
# load config
cfg = yaml.load(open('configs/mb1_120x120.yml'), Loader=yaml.SafeLoader)
# Init FaceBoxes and TDDFA, recommend using onnx flag
onnx_flag = True # or True to use ONNX to speed up
if onnx_flag:
import os
os.environ['KMP_DUPLICATE_LIB_OK'] = 'True'
os.environ['OMP_NUM_THREADS'] = '4'
from FaceBoxes.FaceBoxes_ONNX import FaceBoxes_ONNX
from TDDFA_ONNX import TDDFA_ONNX
face_boxes = FaceBoxes_ONNX()
tddfa = TDDFA_ONNX(**cfg)
else:
face_boxes = FaceBoxes()
tddfa = TDDFA(gpu_mode=False, **cfg)
def inference (img, radio_option):
#set dense_flag from the radiobox option
dense_flag = radio_option in ('2d_dense', '3d', 'depth', 'pncc', 'uv_tex', 'ply', 'obj')
new_suffix = f'.{radio_option}' if radio_option in ('ply', 'obj') else '.jpg'
wfp = f'examples/results/output_{radio_option}_'+''.join(random.choices(string.ascii_uppercase + string.digits, k=5)) + new_suffix
# face detection
boxes = face_boxes(img)
# regress 3DMM params
param_lst, roi_box_lst = tddfa(img, boxes)
# reconstruct vertices
ver_lst = tddfa.recon_vers(param_lst, roi_box_lst, dense_flag=dense_flag)
#RGB to BGR
imgBGR= cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
#decide output according to radiobox arguments
if radio_option == '2d_sparse':
draw_landmarks(imgBGR, ver_lst, show_flag=False, dense_flag=dense_flag, wfp=wfp)
elif radio_option == '2d_dense':
draw_landmarks(imgBGR, ver_lst, show_flag=False, dense_flag=dense_flag, wfp=wfp)
elif radio_option == '3d':
render(img, ver_lst, tddfa.tri, alpha=0.6, show_flag=False, wfp=wfp)
elif radio_option == 'depth':
# if `with_bf_flag` is False, the background is black
depth(imgBGR, ver_lst, tddfa.tri, show_flag=False, wfp=wfp, with_bg_flag=True)
elif radio_option == 'pncc':
pncc(imgBGR, ver_lst, tddfa.tri, show_flag=False, wfp=wfp, with_bg_flag=True)
elif radio_option == 'uv_tex':
uv_tex(imgBGR, ver_lst, tddfa.tri, show_flag=False, wfp=wfp)
elif radio_option == 'pose':
viz_pose(imgBGR, param_lst, ver_lst, show_flag=False,wfp=wfp)
elif radio_option == 'ply':
ser_to_ply(ver_lst, tddfa.tri, height=img.shape[0], wfp=wfp)
elif radio_option == 'obj':
ser_to_obj(imgBGR, ver_lst, tddfa.tri, height=img.shape[0], wfp=wfp)
else:
raise ValueError(f'Unknown opt {radio_option}')
#fetch processed image
if radio_option in ('2d_sparse', '2d_dense', '3d', 'depth', 'pncc', 'uv_tex'):
outimg = cv2.imread(wfp)
resized_img = cv2.resize(outimg, (img.shape[1],img.shape[0]))
resized_img = cv2.cvtColor(resized_img, cv2.COLOR_BGR2RGB)
else:
#no output image(.obj or .ply)
resized_img = img
#if dense_flag is false recalc ver_list with dense flag up
if dense_flag == False:
ver_lst = tddfa.recon_vers(param_lst, roi_box_lst, dense_flag=True)
return render(resized_img, ver_lst, tddfa.tri, alpha=0.6, show_flag=False);
title = "3DDFA V2"
description = "demo for 3DDFA V2. To use it, simply upload your image, or click one of the examples to load them. Read more at the links below."
article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2009.09960'>Towards Fast, Accurate and Stable 3D Dense Face Alignment</a> | <a href='https://github.com/cleardusk/3DDFA_V2'>Github Repo</a></p>"
examples = [
['solvay.jpg'],
['examples/inputs/emma.jpg'],
['examples/inputs/JianzhuGuo.jpg'],
['examples/inputs/trump_hillary.jpg']
]
gr.Interface(
inference,
[
gr.Image(type="numpy", label="Input"),
gr.Radio(['2d_sparse', '2d_dense', '3d', 'depth', 'pncc', 'uv_tex', 'ply', 'obj'], value='2d_sparse')
],
gr.Image(type="numpy", label="Output"),
title=title,
description=description,
article=article,
examples=examples
).launch()
I'm by no means a python expert just managed to work it out from the official docs and APIs. Any optimization/modification would be welcomed.