Object Detection API - EfficientDet D2 - suddenly get OOM on some thousandths iteration!

Train EfficientDet D2 on my own dataset with 10 labels.
I started to get OOM on some thousandths iteration. Last time is was on 54600 iteration!
I made many changes in pipeline.config, to matched the model on my dataset specifications. But the issue of suddenly OOM starts after adding more aspect_ratios.

Train the models on GPU 3090 with docker.
Resize images before convert them to tfrecord.
Batch Size: 8 (max for this GPU)

System Description:

• Linux Ubuntu 20.04
• GeForce 3090
• TF2 - 2.4.3

Configuration (pipeline.config):
model {
ssd {
num_classes: 10
image_resizer {
keep_aspect_ratio_resizer {
min_dimension: 768
max_dimension: 768
pad_to_max_dimension: true
}
}
feature_extractor {
type: "ssd_efficientnet-b2_bifpn_keras"
conv_hyperparams {
regularizer {
l2_regularizer {
weight: 4e-05
}
}
initializer {
truncated_normal_initializer {
mean: 0.0
stddev: 0.03
}
}
activation: SWISH
batch_norm {
decay: 0.99
scale: true
epsilon: 0.001
}
force_use_bias: true
}
bifpn {
min_level: 3
max_level: 7
num_iterations: 5
num_filters: 112
}
}
box_coder {
faster_rcnn_box_coder {
y_scale: 1.0
x_scale: 1.0
height_scale: 1.0
width_scale: 1.0
}
}
matcher {
argmax_matcher {
matched_threshold: 0.5
unmatched_threshold: 0.5
ignore_thresholds: false
negatives_lower_than_unmatched: true
force_match_for_each_row: true
use_matmul_gather: true
}
}
similarity_calculator {
iou_similarity {
}
}
box_predictor {
weight_shared_convolutional_box_predictor {
conv_hyperparams {
regularizer {
l2_regularizer {
weight: 4e-05
}
}
initializer {
random_normal_initializer {
mean: 0.0
stddev: 0.01
}
}
activation: SWISH
batch_norm {
decay: 0.99
scale: true
epsilon: 0.001
}
force_use_bias: true
}
depth: 112
num_layers_before_predictor: 3
kernel_size: 3
class_prediction_bias_init: -4.6
use_depthwise: true
}
}
anchor_generator {
multiscale_anchor_generator {
min_level: 3
max_level: 7
anchor_scale: 4.0
aspect_ratios: 0.25
aspect_ratios: 0.5
aspect_ratios: 0.7
aspect_ratios: 0.85
aspect_ratios: 0.1
aspect_ratios: 1.5
aspect_ratios: 2.0
scales_per_octave: 3
}
}
post_processing {
batch_non_max_suppression {
score_threshold: 1e-08
iou_threshold: 0.5
max_detections_per_class: 100
max_total_detections: 100
}
score_converter: SIGMOID
}
normalize_loss_by_num_matches: true
loss {
localization_loss {
weighted_smooth_l1 {
}
}
classification_loss {
weighted_sigmoid_focal {
gamma: 3.0
alpha: 0.15
}
}
classification_weight: 1.0
localization_weight: 1.0
}
encode_background_as_zeros: true
normalize_loc_loss_by_codesize: true
inplace_batchnorm_update: true
freeze_batchnorm: false
add_background_class: false
}
}
train_config {
batch_size: 8
data_augmentation_options {
random_horizontal_flip {
}
}
data_augmentation_options {
random_scale_crop_and_pad_to_square {
output_size: 768
scale_min: 0.10000000149011612
scale_max: 2.0
}
}
sync_replicas: true
optimizer {
momentum_optimizer {
learning_rate {
cosine_decay_learning_rate {
learning_rate_base: 0.002412
total_steps: 300000
warmup_learning_rate: 0.0000301500021059532
warmup_steps: 2500
}
}
momentum_optimizer_value: 0.9
}
use_moving_average: false
}
fine_tune_checkpoint: "local_path/efficientdet_d2_coco17_tpu-32/checkpoint/ckpt-0"
num_steps: 300000
startup_delay_steps: 0.0
replicas_to_aggregate: 8
max_number_of_boxes: 100
unpad_groundtruth_tensors: false
fine_tune_checkpoint_type: "detection"
use_bfloat16: false
fine_tune_checkpoint_version: V2
}
train_input_reader {
label_map_path: "path_to_workspace/training/annotations/label_map.pbtxt"
tf_record_input_reader {
input_path: "path_to_workspace/training/annotations/train.record"
}
}
eval_config {
metrics_set: "coco_detection_metrics"
use_moving_averages: false
include_metrics_per_category: true
batch_size: 1
}
eval_input_reader {
label_map_path: "path_to_workspace/training/annotations/label_map.pbtxt"
shuffle: false
num_epochs: 1
tf_record_input_reader {
input_path: "/home/tensorflow/models/research/workspace_L/training/annotations/test.record"
}
}

Last OOM Error message:

INFO:tensorflow:Step 54600 per-step time 1.096s loss=0.116
I0726 09:22:30.466438 140430122673984 model_lib_v2.py:702] Step 54600 per-step time 1.096s loss=0.116

Traceback (most recent call last):
File "model_main_tf2.py", line 132, in
tf.compat.v1.app.run()
File "/usr/local/lib/python3.8/dist-packages/tensorflow/python/platform/app.py", line 40, in run
_run(main=main, argv=argv, flags_parser=_parse_flags_tolerate_undef)
File "/usr/local/lib/python3.8/dist-packages/absl/app.py", line 300, in run
_run_main(main, args)
File "/usr/local/lib/python3.8/dist-packages/absl/app.py", line 251, in _run_main
sys.exit(main(argv))
File "model_main_tf2.py", line 123, in main
model_lib_v2.train_loop(
File "/home/tensorflow/.local/lib/python3.8/site-packages/object_detection/model_lib_v2.py", line 690, in train_loop
loss = _dist_train_step(train_input_iter)
File "/usr/local/lib/python3.8/dist-packages/tensorflow/python/eager/def_function.py", line 828, in call
result = self._call(*args, **kwds)
File "/usr/local/lib/python3.8/dist-packages/tensorflow/python/eager/def_function.py", line 855, in _call
return self._stateless_fn(*args, **kwds) # pylint: disable=not-callable
File "/usr/local/lib/python3.8/dist-packages/tensorflow/python/eager/function.py", line 2942, in call
return graph_function._call_flat(
File "/usr/local/lib/python3.8/dist-packages/tensorflow/python/eager/function.py", line 1918, in _call_flat
return self._build_call_outputs(self._inference_function.call(
File "/usr/local/lib/python3.8/dist-packages/tensorflow/python/eager/function.py", line 555, in call
outputs = execute.execute(
File "/usr/local/lib/python3.8/dist-packages/tensorflow/python/eager/execute.py", line 59, in quick_execute
tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name,
tensorflow.python.framework.errors_impl.ResourceExhaustedError: 2 root error(s) found.
(0) Resource exhausted: OOM when allocating tensor with shape[100,257796] and type float on /job:localhost/replica:0/task:0/device:GPU:0 by allocator GPU_0_bfc
[[node Loss/Compare_7/IOU/Intersection/Minimum_1 (defined at home/tensorflow/.local/lib/python3.8/site-packages/object_detection/core/box_list_ops.py:257) ]]
Hint: If you want to see a list of allocated tensors when OOM happens, add report_tensor_allocations_upon_oom to RunOptions for current allocation info.

 [[train_input_images/write_summary/summary_cond/pivot_t/_64/_375]]

Hint: If you want to see a list of allocated tensors when OOM happens, add report_tensor_allocations_upon_oom to RunOptions for current allocation info.

(1) Resource exhausted: OOM when allocating tensor with shape[100,257796] and type float on /job:localhost/replica:0/task:0/device:GPU:0 by allocator GPU_0_bfc
[[node Loss/Compare_7/IOU/Intersection/Minimum_1 (defined at home/tensorflow/.local/lib/python3.8/site-packages/object_detection/core/box_list_ops.py:257) ]]
Hint: If you want to see a list of allocated tensors when OOM happens, add report_tensor_allocations_upon_oom to RunOptions for current allocation info.

0 successful operations.
0 derived errors ignored. [Op:__inference__dist_train_step_85689]

Errors may have originated from an input operation.
Input Source operations connected to node Loss/Compare_7/IOU/Intersection/Minimum_1:
Loss/Compare_7/IOU/Intersection/split (defined at home/tensorflow/.local/lib/python3.8/site-packages/object_detection/core/box_list_ops.py:250)

Input Source operations connected to node Loss/Compare_7/IOU/Intersection/Minimum_1:
Loss/Compare_7/IOU/Intersection/split (defined at home/tensorflow/.local/lib/python3.8/site-packages/object_detection/core/box_list_ops.py:250)

Function call stack:
_dist_train_step -> _dist_train_step

Any ideas why it happens ????
Thank you and sorry if it is too long.