erogol / beauty.torch

Understanding facial beauty with deep learning.

Double Decoder Consistency Audio Samples

This page accompanies this blog post

beauty.torch's Issues

criterion

Hi,
I cannot trace the code to find out what kind of criterion you have employed for training phase.
To be more specific, I am trying to finetune your model on my own data, However I'm encountering the following error, which is related to criterion part:

/home/amir/torch/install/share/lua/5.1/nn/THNN.lua:110: bad argument #2 to 'v' (mismatch between the batch size of input (1) and that of target (2) at /tmp/luarocks_cunn-scm-1-7242/cunn/lib/THCUNN/generic/ClassNLLCriterion.cu:41)

I checked the size of input and target they are the same.

Thank you in advance.

I have met the error "module 'models/init' not found"

Hi, @erogol

I have met the error "module 'models/init' not found"

My error message here:

module 'models/init' not found:No LuaRocks module found for models/initmodule 'models/init' not found:No LuaRocks module found for models/init

OS: Ubuntu14.04
Torch7

Is there any Pytorch version for this beauty net?

Hi, @erogol

Is there any Pytorch version for this beauty net?

Thanks in advance.

from @bemoregt

_pickle.UnpicklingError: invalid load key

model_best.t7

---------------------------------------------------------------------------
UnpicklingError                           Traceback (most recent call last)
<ipython-input-2-7e4eb97b904e> in <module>()
      2 fdet = AlignDlib()
      3 # define scoring model
----> 4 net = HotOrNotTorchRegression(gpu_mode=1,model_path='trained/', model_file='model_best.t7', image_size=224)

~/doc/beauty.torch/utils/Classifier.py in __init__(self, gpu_mode, model_path, model_file, image_size, mean, std)
    162     """
    163     def __init__(self, gpu_mode, model_path, model_file, image_size=224, mean=None, std=None):
--> 164         super(HotOrNotTorchRegression, self).__init__(gpu_mode, model_path=model_path, model_file=model_file, image_size=image_size, mean=mean, std=std, classification=False)
    165 
    166     def classify_image(self, img, N=10):

~/doc/beauty.torch/utils/Classifier.py in __init__(self, gpu_mode, model_path, model_file, image_size, mean, std, classification)
     81         else:
     82             model_path = os.path.join(ROOT_PATH, model_file)
---> 83         self.model = torch.load(model_path)
     84 
     85         if self.classification:

/usr/local/lib/python3.5/dist-packages/torch/serialization.py in load(f, map_location, pickle_module)
    229         f = open(f, 'rb')
    230     try:
--> 231         return _load(f, map_location, pickle_module)
    232     finally:
    233         if new_fd:

/usr/local/lib/python3.5/dist-packages/torch/serialization.py in _load(f, map_location, pickle_module)
    367 
    368     f.seek(0)
--> 369     magic_number = pickle_module.load(f)
    370     if magic_number != MAGIC_NUMBER:
    371         raise RuntimeError("Invalid magic number; corrupt file?")

UnpicklingError: invalid load key, '�'.

How that might be a problem?

Fails in CUDA

Finally I've managed to install all prerequisites (cunn, CUDA, cutorch)
Now I try to run an example as in your notebook:

import lutorpy as lua
import numpy as np
from skimage import io
from utils.img_processing import process_img
from utils.face_dlib import AlignDlib
from utils.Classifier import HotOrNotTorchRegression

# define face model
fdet = AlignDlib()
# define scoring model
net = HotOrNotTorchRegression(gpu_mode=0,model_path='trained/', model_file='model_best.t7', image_size=224)

# score function
def score_image(img):
    # find all faces
    dets = fdet.getAllFaceBoundingBoxes(img)
    # if there is only one match, go and score
    if len(dets) != 0:
        det = dets[0]
        img2, face_img = process_img(img, fdet, size=224, is_shortest=True,
                                    margins={'width':0.5, "bottom":0.5, "top":0.5},
                                    bbox={'x1':int(np.round(det.left())),
                                           'x2':int(np.round(det.right())),
                                           'y1':int(np.round(det.top())),
                                           'y2':int(np.round(det.bottom()))})
        score =  net.classify_image(img2)[1]
        return img2, score
    else:
        return None, None

# read an example image from web
img_url = "https://s-media-cache-ak0.pinimg.com/originals/9a/3b/65/9a3b65c368c712ca322394ee4d19d407.jpg"
img = io.imread(img_url)
score = score_image(img)[1]

Output:

THCudaCheck FAIL file=/home/serj/torch/extra/cutorch/lib/THC/THCGeneral.c line=66 error=30 : unknown error
Traceback (most recent call last):
  File "example.py", line 6, in <module>
    from utils.Classifier import HotOrNotTorchRegression
  File "/home/serj/work/beautytorch/beauty.torch/utils/Classifier.py", line 11, in <module>
    lua.require('cunn')
  File "/home/serj/work/beautytorch/venv/local/lib/python2.7/site-packages/lutorpy/__init__.py", line 112, in require
    ret = luaRuntime.require(module_name)
  File "lutorpy/_lupa.pyx", line 318, in lutorpy._lupa.LuaRuntime.require (lutorpy/_lupa.c:6189)
  File "lutorpy/_lupa.pyx", line 1658, in lutorpy._lupa.call_lua (lutorpy/_lupa.c:25877)
  File "lutorpy/_lupa.pyx", line 1667, in lutorpy._lupa.execute_lua_call (lutorpy/_lupa.c:25987)
  File "lutorpy/_lupa.pyx", line 1620, in lutorpy._lupa.raise_lua_error (lutorpy/_lupa.c:25297)
lutorpy._lupa.LuaError: cuda runtime error (30) : unknown error at /home/serj/torch/extra/cutorch/lib/THC/THCGeneral.c:66

Maybe the problem is that my GPU is not NVidia. Can I disable cunn somehow? Please help!

ADD: can't run convert2cpu because of the same.

UPD As adviced by cutorch developers the conversion from GPU to CPU model is not possible without having NVidia GPU. It would be very helpful if CPU model could be shared as well.

How can I get the dataset?

How can I get the dataset?