用TensorFlow的Object Detection API做目标识别，一直能正常训练、导出模型、加载训练好的模型、在图片里有效搜索识别。
以前训练好的模型现在还都正常（再次导出、加载、识别均正常）。
现在新训练的模型和以前一样的环境：python 3.9,tensorflow==2.8.0。
新训练的也能导出模型，加载训练好的模型就出错，就错在这一句：
detect_fn = tf.saved_model.load(PATH_TO_SAVED_MODEL)
改成detect_fn = tf.saved_model.load(PATH_TO_SAVED_MODEL, tags=None, options=None)还是出错。
报错好像是：恢复检查点有问题。
我故意把checkpoint文件第一行里model_checkpoint_path: "ckpt-11"改成10，导出模型后还是加载不了模型。
怎么会非得要shape (3, 3, 256, 12)和shape (3, 3, 256, 24)兼容呢？
老板找事，实在是没办法，束手无策。
完整错误信息如下：
Traceback (most recent call last):
File "C:\TensorFlowObjectDetectionAPITest\Bill\inference_main_v2.py", line 73, in <module>
detect_fn = tf.saved_model.load(PATH_TO_SAVED_MODEL)
File "C:\ProgramData\Anaconda3\envs\tensorflow2\lib\site-packages\tensorflow\python\saved_model\load.py", line 782, in load
result = load_partial(export_dir, None, tags, options)["root"]
File "C:\ProgramData\Anaconda3\envs\tensorflow2\lib\site-packages\tensorflow\python\saved_model\load.py", line 912, in load_partial
loader = Loader(object_graph_proto, saved_model_proto, export_dir,
File "C:\ProgramData\Anaconda3\envs\tensorflow2\lib\site-packages\tensorflow\python\saved_model\load.py", line 189, in __init__
self._restore_checkpoint()
File "C:\ProgramData\Anaconda3\envs\tensorflow2\lib\site-packages\tensorflow\python\saved_model\load.py", line 507, in _restore_checkpoint
load_status = saver.restore(variables_path, self._checkpoint_options)
File "C:\ProgramData\Anaconda3\envs\tensorflow2\lib\site-packages\tensorflow\python\training\tracking\util.py", line 1468, in restore
base.CheckpointPosition(
File "C:\ProgramData\Anaconda3\envs\tensorflow2\lib\site-packages\tensorflow\python\training\tracking\base.py", line 295, in restore
restore_ops = trackable._restore_from_checkpoint_position(self) # pylint: disable=protected-access
File "C:\ProgramData\Anaconda3\envs\tensorflow2\lib\site-packages\tensorflow\python\training\tracking\base.py", line 1060, in _restore_from_checkpoint_position
current_position.checkpoint.restore_saveables(tensor_saveables,
File "C:\ProgramData\Anaconda3\envs\tensorflow2\lib\site-packages\tensorflow\python\training\tracking\util.py", line 349, in restore_saveables
new_restore_ops = functional_saver.MultiDeviceSaver(
File "C:\ProgramData\Anaconda3\envs\tensorflow2\lib\site-packages\tensorflow\python\training\saving\functional_saver.py", line 415, in restore
restore_ops = restore_fn()
File "C:\ProgramData\Anaconda3\envs\tensorflow2\lib\site-packages\tensorflow\python\training\saving\functional_saver.py", line 398, in restore_fn
restore_ops.update(saver.restore(file_prefix, options))
File "C:\ProgramData\Anaconda3\envs\tensorflow2\lib\site-packages\tensorflow\python\training\saving\functional_saver.py", line 112, in restore
restore_ops[saveable.name] = saveable.restore(
File "C:\ProgramData\Anaconda3\envs\tensorflow2\lib\site-packages\tensorflow\python\training\saving\saveable_object_util.py", line 133, in restore
raise ValueError(
ValueError: Received incompatible tensor with shape (3, 3, 256, 12) when attempting to restore variable with shape (3, 3, 256, 24) and name _model/_box_predictor/variables/2/.ATTRIBUTES/VARIABLE_VALUE.
PATH_TO_SAVED_MODEL,变量没定义？
如果不是上一位学者说的情况，就可能是保存模型是出错，导致参数量不一致，你可以再检查下，具体可以参考https://blog.csdn.net/weixin_42416791/article/details/120085728。