wellyzhang / raven Goto Github PK

Hi, I am a little bit confused about what does the Progression 'relation' of Position 'attribute' mean. In the 1st example on page 5 of supp material, it says there is a progression relation on position, but I don't understand why is this the case. The position does not seem to change in a uniform pattern. Thanks if you could answer me.

Hello @WellyZhang,

Thanks for making the code public. I used all the default hyperparameters. The environment is Python 2.7, Pytorch 0.4.1, Torchvision 0.2.0. I stopped training at 186 epochs. The best test accuracy I am able to get is 62.9% more than 59.56%. Were you able to get more accuracy than reported in the paper? Is it possible I made somewhere something wrong? If possible could you provide your logs or best models? Thanks and Regards

Thanks in advance

I want to get the dataset as an image file format like .png or .jpeg. However, after loading .npz file, the image of the data does not look good. Here is the example of the image I got. After running the codes below, the image looks like that. Could you please tell me how to normalize the image?

data = np.load('RAVEN_5_train.npz')
image = data['image']
print(image.shape, image.dtype)
img = Image.fromarray(image, 'RGB')
plt.figure()
plt.imshow(img)
plt.show()

command line run of:
python src/model/main.py --model CNN_LSTM --path Data --device -1
returns error.
Error occurs in forward model pass with "RuntimeError: shape '[-1, 1, 80, 80]' is invalid for input of size 25690112"

A similar error occurs for CNN_MLP

When running python src/dataset/main.py, I get the following error:

Traceback (most recent call last):
File "src/dataset/main.py", line 328, in
main()
File "src/dataset/main.py", line 324, in main
separate(args, all_configs)
File "src/dataset/main.py", line 282, in separate
dom = dom_problem(context + candidates, rule_groups)
File "/home/aarchan/RAVEN/src/dataset/serialize.py", line 106, in dom_problem
entity_l.set("mask", rle_encode(get_mask(entity_bbox, entity_type, entity_size, entity_angle)))
File "/home/aarchan/RAVEN/src/dataset/api.py", line 90, in rle_encode
runs[1::2] -= runs[::2]
ValueError: operands could not be broadcast together with shapes (225,) (226,) (225,)

Seems like this error occurs when the number of elements in runs is odd, causing runs[1::2] and runs[::2] to have different numbers of elements. Instead, should this line be runs[1::2] -= runs[:-1:2], if the number of elements in runs is odd?

Could you please help me fix this issue? Thanks!