mindspore.ops.Concat
【操作步骤&问题现象】
1、输入tensor变多之后 concat会报错
代码：（rois shape变成(6,5)就会报错）
def construct(self, input_cls, input_reg, rois): cls_vector = self.reshape(self.roi_align(input_cls, rois), (rois.shape[0], -1)) reg_vector = self.reshape(self.roi_align(input_reg, rois), (rois.shape[0], -1)) # (N, 196, 7, 7) roi_cls_score = () roi_reg_score = () for per_roi in range(rois.shape[0]): cls_score = cls_vector[per_roi, 0::self.reg_stride].sum()/self.classes_num for per_class in range(1, self.classes_num): cls_score = mnp.append(cls_score, cls_vector[per_roi, per_class * self.bins_num::self.cls_stride].sum()/self.classes_num) reg_score = reg_vector[per_roi, 0::self.stride].sum() / 4 for per_axis in range(1,4): reg_score = mnp.append(reg_score, reg_vector[per_roi, per_axis * self.bins_num::self.reg_stride].sum()/4) roi_cls_score += (self.softmax(cls_score),) roi_reg_score += (reg_score,) roi_cls_score = self.concat(roi_cls_score) roi_reg_score = self.concat(roi_reg_score) return roi_cls_score, roi_reg_score import mindsporefrom mindspore import context context.set_context(mode=context.GRAPH_MODE, device_target="Ascend", device_id=4)# roi = PSRoIAlign(7,7,0.0625,2) input_cls = Tensor(np.random.rand(1, 3969, 80, 38), mindspore.float32) input_reg = Tensor(np.random.rand(1, 196, 80, 38), mindspore.float32) rois = Tensor(np.array([[0,1,1,2,2],[1,30,30,50,50],[2,30,30,50,50],[3,30,30,50,50]]), mindspore.float32) //rois shape变成(6,5)就会报错 cls_score_map, reg_score_map = roi(input_cls,input_reg, rois) # (N, 81*7*7, 7, 7)print(cls_score_map.shape)print(reg_score_map.shape)
rois shape为(5,5) 时返回 但(6,5)就会报错
具体报错请看附件
(ci3.7) [root@ascend90 fr]# python ps_roi_pool.py [ERROR] ANALYZER(7820,ffff8983f630,python):2021-10-25-22:09:52.345.962 [mindspore/ccsrc/pipeline/jit/static_analysis/evaluator.cc:77] EnterStackFrame] Exceed function call depth limit 1000, (function call depth: 1001, simulate call depth: 999), please call 'context.set_context(max_call_depth=value)' to adjust this value. [ERROR] DEBUG(7820,ffff8983f630,python):2021-10-25-22:09:52.346.059 [mindspore/ccsrc/debug/trace.cc:118] TraceGraphEval] *******************************graph evaluate stack********************************** [ERROR] DEBUG(7820,ffff8983f630,python):2021-10-25-22:09:52.375.526 [mindspore/ccsrc/debug/trace.cc:122] TraceGraphEval]
解答：
[ERROR] ANALYZER(7820,ffff8983f630,python):2021-10-25-22:09:52.345.962 [mindspore/ccsrc/pipeline/jit/static_analysis/evaluator.cc:77] EnterStackFrame] Exceed function call depth limit 1000, (function call depth: 1001, simulate call depth: 999), please call 'context.set_context(max_call_depth=value)' to adjust this value.
报错提示：超过函数调用深度限制1000，(函数调用深度:1001，模拟调用深度:999)，请调用context.set_context(max_call_depth=value)来调整这个值。
