noagarcia / keras_rmac Goto Github PK

Hi,This model Spend too much time,How to solve?

when i load weight vgg16 , it happend issue "Dimension 0 in both shapes must be equal, but are 344064 and 25088"

When I load the pre-trained model, I got dimension mismatched error, when I resized images to (224, 224), it was solved. Does it affect the model? I mean r-mac model.

x = image.img_to_array(img) #把图像转换为数组类形numpy.ndarray
x = np.expand_dims(x, axis=0)
x = utils.preprocess_image(x)
The x output shape here is (1, 683, 1024, 3)
but x = utils.preprocess_image(x)
x[:, 0, :, :] -= 103.939
x[:, 1, :, :] -= 116.779
x[:, 2, :, :] -= 123.68
The second dimension is the number of channels, but the last dimension of the shape of x is the number of channels

I have tried that loading VGG16 weight only at the beginning of program, and every time after executing ,I'll delete the loading image. But it still memory leak, how can I release my memory? (not GPU Ram)

Hi,

can you please tell me which is the dataset used to generate the PCA?

I am using theano as backend. Is there a way to use GPU instead of the CPU for processing..
And plus is there a way to run multiple processes as there is a lock while execution.

The region array calculated by rmac_regions seems to be [x, y, w, h], but actually, it is used as [x, y, h, w] in RoiPooling...

I ran your code in my local system and I got the following error,
Traceback (most recent call last): File "rmac.py", line 92, in <module> model = rmac((x.shape[1], x.shape[2], x.shape[3]), len(regions)) File "rmac.py", line 34, in rmac vgg16_model = VGG16(utils.DATA_DIR + utils.WEIGHTS_FILE, input_shape) File "/home/randheer/keras_rmac/vgg16.py", line 61, in VGG16 model.load_weights(weights_path) File "/home/randheer/anaconda3/lib/python3.6/site-packages/keras/engine/network.py", line 1166, in load_weights f, self.layers, reshape=reshape) File "/home/randheer/anaconda3/lib/python3.6/site-packages/keras/engine/saving.py", line 1058, in load_weights_from_hdf5_group K.batch_set_value(weight_value_tuples) File "/home/randheer/anaconda3/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py", line 2465, in batch_set_value assign_op = x.assign(assign_placeholder) File "/home/randheer/anaconda3/lib/python3.6/site-packages/tensorflow/python/ops/variables.py", line 1762, in assign name=name) File "/home/randheer/anaconda3/lib/python3.6/site-packages/tensorflow/python/ops/state_ops.py", line 223, in assign validate_shape=validate_shape) File "/home/randheer/anaconda3/lib/python3.6/site-packages/tensorflow/python/ops/gen_state_ops.py", line 64, in assign use_locking=use_locking, name=name) File "/home/randheer/anaconda3/lib/python3.6/site-packages/tensorflow/python/framework/op_def_library.py", line 788, in _apply_op_helper op_def=op_def) File "/home/randheer/anaconda3/lib/python3.6/site-packages/tensorflow/python/util/deprecation.py", line 507, in new_func return func(*args, **kwargs) File "/home/randheer/anaconda3/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 3300, in create_op op_def=op_def) File "/home/randheer/anaconda3/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1823, in __init__ control_input_ops) File "/home/randheer/anaconda3/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1662, in _create_c_op raise ValueError(str(e))
ValueError: Dimension 0 in both shapes must be equal, but are 344064 and 25088. Shapes are [344064,4096] and [25088,4096]. for 'Assign_26' (op: 'Assign') with input shapes: [344064,4096], [25088,4096].

My directory looks like this:
->data
---->vgg_weights.h5
---->sample.jpg
->RoiPooling.py
->rmac.py
->vgg16.py
->get_regions.py
->utils.py

Can you address this problem?
Waiting for a reply

def rmac_regions(W, H, L):#这里L使用的是3
#w=32,h=21
ovr = 0.4 # desired overlap of neighboring regions
steps = np.array([2, 3, 4, 5, 6, 7], dtype=np.float) # possible regions for the long dimension

w = min(W,H) 

b = (max(H,W) - w)/(steps-1)
idx = np.argmin(abs(((w ** 2 - w*b)/w ** 2)-ovr)) # steps(idx) regions for long dimension
# region overplus per dimension
Wd, Hd = 0, 0
if H < W:
    Wd = idx + 1
elif H > W:
    Hd = idx + 1

What do you mean,step, b and idx，Wd，Hd in the code?