yassersouri / fast-bird-part-localization Goto Github PK

The "body" detector training bug log:

fast_bird_part_localization/geometry_utils.py:59: RuntimeWarning: overflow encountered in ubyte_scalars
cenX = float(self.xmin + self.xmax) / 2
fast_bird_part_localization/geometry_utils.py:60: RuntimeWarning: overflow encountered in ubyte_scalars
cenY = float(self.ymin + self.ymax) / 2
done 0 in 0:00:00.000088
done 100 in 0:07:55.448850
...
done 2400 in 1:50:29.576296
done 2500 in 1:55:50.305564
Traceback (most recent call last):
File "create_rf_model.py", line 65, in
main()
File "create_rf_model.py", line 36, in main
pos, neg = dh.image_point_features(img, part_box, part_name)
File "fast_bird_part_localization/cnn_utils.py", line 108, in image_point_features
return self.features(positive_points), self.features(negative_points)
File "fast_bird_part_localization/cnn_utils.py", line 91, in features
features[i, :] = self.ffeats[point[0], point[1], :]
IndexError: index 341 is out of bounds for axis 0 with size 333

flatten makes a copy of the data

The 'bbox' detector training bug: if one changes the part name from 'head' to 'bbox' at
https://github.com/yassersouri/fast-bird-part-localization/blob/master/create_rf_model.py#L23

However, because the parameters setting dict has no key-value pair for 'bbox' at:
https://github.com/yassersouri/fast-bird-part-localization/blob/master/fast_bird_part_localization/settings.py#L10

So the training for 'bbox' will fail at:
https://github.com/yassersouri/fast-bird-part-localization/blob/master/fast_bird_part_localization/cnn_utils.py#L103

Hi,

Nice code! I was wondering which type of split function is used by scikit-learn library, any idea? I am not sure if typically used value/pixel difference split function will work well with hyper-column features, since there is no correlation between feature values.

Thanks!

after running "head, head_prob = fbp.detect(img)", I have got these errors:

head, head_prob = fbp.detect(img)
Traceback (most recent call last):
File "", line 1, in
File "fast_bird_part_localization/detector.py", line 23, in detect
pred_box = geometry_utils.find_rect_from_preds(geometry_utils.post_process_preds(preds_img))
File "fast_bird_part_localization/geometry_utils.py", line 205, in post_process_preds
preds = skimage.morphology.closing(preds, skimage.morphology.square(10))
File "/usr/lib/python2.7/dist-packages/skimage/morphology/grey.py", line 215, in closing
dilated = dilation(image, selem)
File "/usr/lib/python2.7/dist-packages/skimage/morphology/grey.py", line 112, in dilation
image = img_as_ubyte(image)
File "/usr/lib/python2.7/dist-packages/skimage/util/dtype.py", line 358, in img_as_ubyte
return convert(image, np.uint8, force_copy)
File "/usr/lib/python2.7/dist-packages/skimage/util/dtype.py", line 99, in convert
raise ValueError("can not convert %s to %s." % (dtypeobj_in, dtypeobj))
ValueError: can not convert int64 to uint8.