Do you have a 3Ddata of crop_and_resize function implementation? DATA of shape [batch, dim1, dim2, dim3, channel]. If not, how to implement this function in tensorflow.
On 3D data, I deduct the corresponding area of each box from the feature graph. The number of BOX of an input image is unknown, the dimension of box is [None, cord-dim], and the number of boxes is None, because the input data is placed holder. So it is not feasible to use two-dimensional crop_and_resize functions.
# labels_mask_box: N * 6
roi_align(tf.expand_dims(netconv4[i, ...], axis=0), labels_mask_box[..., :] , crop_szie)
tf.image.crop_and_resize(image, boxes, box_ind,crop_size=[crop_size, crop_size])