Please make sure that this is a bug. As per our
GitHub Policy,
we only address code/doc bugs, performance issues, feature requests and
build/installation issues on GitHub. tag:bug_template

System information

• Have I written custom code (as opposed to using a stock example script provided in TensorFlow): Yes
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Ubuntu 18.04
• TensorFlow installed from (source or binary): Binary
• TensorFlow version (use command below): TF:2.5.0-dev20210114
• Python version: 3.7
• CUDA/cuDNN version: 11.0, 8.0.4
• GPU model and memory: 1060

Describe the current behavior
TensorRT converter crashes with a segmentation fault when I try to export my saved_model.
Interestingly, if I set minimum_segment_size=10, it works because it skips

Replaced segment 5 consisting of 7 nodes by StatefulPartitionedCall/decode_predictions/TRTEngineOp_0_5.
2021-01-15 15:21:38.915310: I tensorflow/compiler/tf2tensorrt/convert/convert_graph.cc:858] Segment consists of nodes: StatefulPartitionedCall/decode_predictions/combined_non_max_suppression/CombinedNonMaxSuppression, StatefulPartitionedCall/decode_predictions/combined_non_max_suppression/CombinedNonMaxSuppression/max_output_size_per_class, StatefulPartitionedCall/decode_predictions/combined_non_max_suppression/Const, StatefulPartitionedCall/decode_predictions/combined_non_max_suppression/iou_threshold, StatefulPartitionedCall/decode_predictions/combined_non_max_suppression/score_threshold, StatefulPartitionedCall/decode_predictions/transpose_1, StatefulPartitionedCall/decode_predictions/transpose_1/perm

I have attached the full log after running with these flags
TF_CPP_VMODULE=trt_engine_op=2,convert_nodes=2,convert_graph=2,segment=2,trt_shape_optimization_profiles=2,trt_engine_resource_ops=2 python trt.py

Standalone code to reproduce the issue

import os

import tensorflow as tf

## Download and extract the zip 
## URL: https://drive.google.com/file/d/1Zxqdnm2iHpJGdUl17cAi-lV7wZ3UhMDA/view

params = tf.experimental.tensorrt.ConversionParams(
    precision_mode='FP32',
    maximum_cached_engines=1,
    minimum_segment_size=5)

converter = tf.experimental.tensorrt.Converter(
    input_saved_model_dir='retinanet-18-640-30x-64-tpu',
    conversion_params=params)
converter.convert()

def input_fn(steps=1):
    for i in range(steps):
        yield (tf.random.uniform([640, 640, 3]), tf.constant(1, dtype=tf.int32))
        
converter.build(input_fn=input_fn)
converter.save('trt')

Other info / logs Include any logs or source code that would be helpful to
diagnose the problem. If including tracebacks, please include the full
traceback. Large logs and files should be attached.
trt_log.txt