当我在“ samples / shapes / shapes.py”中运行train_shapes.py时，出现如下错误：
回溯（最近一次通话最近）：文件“ MaskRCNN-Unet / samples / shapes / train_shapes.py”，第274行， 在layers =“ heads”中）文件“ MaskRCNN-Unet / mrcnn / modelxmask.py”，第2844行， 在火车use_multiprocessing = True中，文件“ /usr/local/lib/python3.5/dist-packages/keras/legacy/interfaces.py”，第91行， 在包装器中返回func（* args，** kwargs）文件“ /usr/local/lib/python3.5/dist-packages/keras/engine/training.py”，第1413行， 在fit_generator initial_epoch = initial_epoch中）文件“ /usr/local/lib/python3.5/dist-packages/keras/engine/training_generator.py”，第214行， 在fit_generator class_weight = class_weight中）文件“ /usr/local/lib/python3.5/dist-packages/keras/engine/training.py”，第1213行， 在train_on_batch输出中= self.train_function（ins）文件“ /usr/local/lib/python3.5/dist-packages/keras/backend/tensorflow_backend.py”，行2715，在调用返回self._call（inputs）文件“ /usr/local/lib/python3.5/dist-packages/keras/backend/tensorflow_backend.py“，第2675行， 在_call fetched = self._callable_fn（* array_vals）中，文件“ /usr/local/lib/python3.5/dist-packages/tensorflow/python/client/session.py”，行1439，在调用run_metadata_ptr中）文件“ / usr /local/lib/python3.5/dist-packages/tensorflow/python/framework/errors_impl.py“，第528行， 在出口c_api.TF_GetCode（self.status.status））tensorflow.python.framework.errors_impl.InvalidArgumentError：预期开始，结束和步幅为一维相等大小的张量，但形状为[2,1]，[2,1 ]和[2]代替。 [[{{node DetectSmaskBbox / map / while / strided_slice_1}}]
我认为提示不明确，我无法理解并找到此错误，有人可以帮助我吗？预先感谢。
我试图注释掉代码中的DetectSmaskBbox层，并且它起作用了。因此，我认为DetectSmaskBbox层中一定有问题。
class DetectSmaskBbox(KE.Layer):
def __init__(self, config, mode, **kwargs):
super(DetectSmaskBbox, self).__init__(**kwargs)
self.config = config
self.mode = mode
def call(self, inputs):
input_image_meta, mrcnn_detections, smask = inputs
xmask_smasks = self.detect_smask_bbox(input_image_meta, mrcnn_detections, smask)
return xmask_smasks
def detect_smask_bbox(self, image_metas, detections, smasks):
"""Map the bbox generated by mask-rcnn to smasks.
images: List of image tensors.
Returns a list of tensors. [masks1, mask2, mask3, ...]
masks: [Batch, H, W, N] instance logits masks, different image has different numbers of masks.
"""
instance_count = self.config.DETECTION_MAX_INSTANCES
# Process detections
results = []
m = parse_image_meta_graph(image_metas)
image_shape = m['image_shape'][0]
print("the shape of image_shape is ", image_shape.get_shape().as_list())
original_image_shape = m['original_image_shape'][0]
print("the shape of original_image_shape is ", original_image_shape.get_shape().as_list())
window = m['window'][0]
for i in range(self.config.BATCH_SIZE):
detections_per_img = detections[i]
xmask_smasks = self.unmold_detections_to_smask(detections_per_img, smasks[i], image_shape,
original_image_shape, window)
# xmask_smasks: [num_instances, Height, Width]
# padding
padding = tf.maximum(instance_count - tf.shape(xmask_smasks)[0], 0)
xmask_smasks = tf.pad(xmask_smasks, [(0, padding), (0, 0), (0, 0)]) # xmask_smasks: [instance_count, Height, Width]
t = tf.split(xmask_smasks, instance_count, axis=0) # t is a list of tensors with shape [1, Height, Width]
xmask_smasks = tf.stack(t, axis=-1) # [1, Height, Width, instance_count]
xmask_smasks = tf.squeeze(xmask_smasks, axis=0) # xmask_smasks: [Height, Width, instance_count]
results.append(xmask_smasks)
results = tf.stack(results, axis=0)
return results # results: [Batch, Height, Width, instance_count]
def unmold_detections_to_smask(self, detections, smask, original_image_shape,
image_shape, window):
"""Map the detections of one image that generated by mask-rcnn to smasks, set
the values outside the bbox to be zeros.
detections: [N, (y1, x1, y2, x2, class_id, score)] in normalized coordinates
smask: [height, width]
original_image_shape: [H, W, C] Original image shape before resizing
image_shape: [H, W, C] Shape of the image after resizing and padding
window: [y1, x1, y2, x2] Pixel coordinates of box in the image where the real
image is excluding the padding.
Returns: full_masks of an image
full_masks: [num_instances, Height, Width] Instance masks of smask. Height
and Width equals to the height and width of original images.
"""
# How many detections do we have?
# Detections array is padded with zeros. Find the first class_id == 0.
zero_ix = tf.where(tf.equal(detections[:, 4],0))
# is_not_empty = tf.cond(tf.greater(tf.size(zero_ix), 0), lambda: 1, lambda: 0)
# N = zero_ix[0][0] if is_not_empty else detections.get_shape[0]
N = tf.cond(tf.greater(tf.size(zero_ix), 0), lambda: tf.cast(zero_ix[0][0], tf.int32), lambda: detections.get_shape()[0])
# Extract boxes, class_ids, scores, and class-specific masks
boxes = detections[:N, :4]
# class_ids = tf.cast(detections[:N, 4], tf.int32)
# scores = detections[:N, 5]
# Translate normalized coordinates in the resized image to pixel
# coordinates in the original image before resizing
print("the shape of window is ", window.get_shape().as_list())
print("the shape of image_shape is ", image_shape.get_shape().as_list())
window = norm_boxes_graph(window, image_shape[:2])
wy1, wx1, wy2, wx2 = tf.split(window, 4)
shift = tf.concat([wy1, wx1, wy1, wx1], axis=-1)
wh = wy2 - wy1 # window height
ww = wx2 - wx1 # window width
scale = tf.concat([wh, ww, wh, ww], axis=-1)
# Convert boxes to normalized coordinates on the window
boxes = tf.divide(boxes - shift, scale)
# Convert boxes to pixel coordinates on the original image
boxes = denorm_boxes_graph(boxes, original_image_shape[:2])
# Filter out detections with zero area. Happens in early training when
# network weights are still random
include_ix = tf.where((boxes[:, 2] - boxes[:, 0]) * (boxes[:, 3] - boxes[:, 1]) > 0)
include_ix = tf.squeeze(include_ix)
# is_empty = tf.cond(tf.greater(tf.size(include_ix), 0), lambda: 1, lambda: 0)
# if is_empty is not None:
boxes = tf.gather(boxes, include_ix, axis=0)
# class_ids = tf.gather(class_ids, include_ix, axis=0)
# scores = tf.gather(scores, include_ix, axis=0)
#N = class_ids.shape[0]
# Resize masks to original image size and set boundary threshold.
# full_masks = []
fn = lambda box: utils.map_bbox_to_smask(smask, box, original_image_shape[:2])
full_masks = tf.map_fn(fn, boxes, dtype=tf.float32)
return full_masks
很抱歉，由于我不知道问题出在哪里，所以无法明确问题。但是可以在“ https://github.com/zlyx525/MaskRCNN-UNet”中找到完整的项目。非常感谢！