Here are four RTXA6000(48G) set each processing 4 images GPU memory overflow problem (IMAGE_MIN_DIM = 1024 IMAGE_MAX_DIM = 1024) (Batchsize = 16)

[xxxming730]

First of all, I wish all developers good health and all the best ！

My training is divided into "head" and "all" phase, but at the beginning of training "all" phase, there was a GPU memory overflow error,
I was very surprised why four RTXA6000 (48G) would get GPU memory overflow error with GPU COUNT = 4 IMAGES_PER_GPU = 4 settings,
because mrcnn/config.py describes the IMAGES_PER_GPU parameter as follows:

Number of images to train with on each GPU. A 12GB GPU can typically handle 2 images of 1024x1024px.

Adjust based on your GPU memory and image sizes. Use the highest number that your GPU can handle for best performance.

What causes such a problem?

    My model building environment is:

cuda11.3 + cudnn8.2.1 + tensorflow-gpu2.6 + python3.8

    My config are as follows ：

Configurations:
BACKBONE resnet101
BACKBONE_STRIDES [4, 8, 16, 32, 64]
BATCH_SIZE 16
BBOX_STD_DEV [0.1 0.1 0.2 0.2]
COMPUTE_BACKBONE_SHAPE None
DETECTION_MAX_INSTANCES 100
DETECTION_MIN_CONFIDENCE 0.7
DETECTION_NMS_THRESHOLD 0.3
FPN_CLASSIF_FC_LAYERS_SIZE 1024
GPU_COUNT 4
GRADIENT_CLIP_NORM 5.0
IMAGES_PER_GPU 4
IMAGE_CHANNEL_COUNT 3
IMAGE_MAX_DIM 1024
IMAGE_META_SIZE 22
IMAGE_MIN_DIM 1024
IMAGE_MIN_SCALE 0
IMAGE_RESIZE_MODE square
IMAGE_SHAPE [1024 1024 3]
LEARNING_MOMENTUM 0.9
LEARNING_RATE 0.001
LOSS_WEIGHTS {'rpn_class_loss': 1.0, 'rpn_bbox_loss': 1.0, 'mrcnn_class_loss': 1.0, 'mrcnn_bbox_loss': 1.0, 'mrcnn_mask_loss': 1.0}
MASK_POOL_SIZE 14
MASK_SHAPE [28, 28]
MAX_GT_INSTANCES 100
MEAN_PIXEL [123.7 116.8 103.9]
MINI_MASK_SHAPE (56, 56)
NAME medtest_2464_
NUM_CLASSES 10
POOL_SIZE 7
POST_NMS_ROIS_INFERENCE 1000
POST_NMS_ROIS_TRAINING 2000
PRE_NMS_LIMIT 6000
ROI_POSITIVE_RATIO 0.33
RPN_ANCHOR_RATIOS [0.5, 1, 2]
RPN_ANCHOR_SCALES (32, 64, 128, 256, 512)
RPN_ANCHOR_STRIDE 1
RPN_BBOX_STD_DEV [0.1 0.1 0.2 0.2]
RPN_NMS_THRESHOLD 0.7
RPN_TRAIN_ANCHORS_PER_IMAGE 256
STEPS_PER_EPOCH 308
TOP_DOWN_PYRAMID_SIZE 256
TRAIN_BN False
TRAIN_ROIS_PER_IMAGE 200
USE_MINI_MASK True
USE_RPN_ROIS True
VALIDATION_STEPS 50
WEIGHT_DECAY 0.0001

[xxxming730]

tensorflow.python.framework.errors_impl.ResourceExhaustedError: 5 root error(s) found.
(0) Resource exhausted: OOM when allocating tensor with shape[800,256,28,28] and type float on /job:localhost/replica:0/task:0/device:GPU:0 by allocator GPU_0_bfc
[[{{node training_2/SGD/gradients/gradients/tower_0/mask_rcnn/mrcnn_mask/conv2d_6/Conv2D_grad/Conv2DBackpropInput}}]]
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. This isn't available when running in Eager mode.

 [[tower_2/mask_rcnn/proposal_targets/strided_slice_128/_27411]]

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. This isn't available when running in Eager mode.

(1) Resource exhausted: OOM when allocating tensor with shape[800,256,28,28] and type float on /job:localhost/replica:0/task:0/device:GPU:0 by allocator GPU_0_bfc
[[{{node training_2/SGD/gradients/gradients/tower_0/mask_rcnn/mrcnn_mask/conv2d_6/Conv2D_grad/Conv2DBackpropInput}}]]
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. This isn't available when running in Eager mode.

 [[tower_2/mask_rcnn/proposal_targets/RandomShuffle_7/_27767]]

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. This isn't available when running in Eager mode.

(2) Resource exhausted: OOM when allocating tensor with shape[800,256,28,28] and type float on /job:localhost/replica:0/task:0/device:GPU:0 by allocator GPU_0_bfc
[[{{node training_2/SGD/gradients/gradients/tower_0/mask_rcnn/mrcnn_mask/conv2d_6/Conv2D_grad/Conv2DBackpropInput}}]]
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. This isn't available when running in Eager mode.

 [[tower_1/mask_rcnn/proposal_targets/RandomShuffle_2/_27389]]

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. This isn't available when running in Eager mode.

(3) Resource exhausted: OOM when allocating tensor with shape[800,256,28,28] and type float on /job:localhost/replica:0/task:0/device:GPU:0 by allocator GPU_0_bfc
[[{{node training_2/SGD/gradients/gradients/tower_0/mask_rcnn/mrcnn_mask/conv2d_6/Conv2D_grad/Conv2DBackpropInput}}]]
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. This isn't available when running in Eager mode.

 [[tower_3/mask_rcnn/proposal_targets/RandomShuffle_7/_27775]]

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. This isn't available when running in Eager mode.

(4) Resource exhausted: OOM when allocating tensor with shape[800,256,28,28] and type float on /job:localhost/replica:0/task:0/device:GPU:0 by allocator GPU_0_bfc
[[{{node training_2/SGD/gradients/gradients/tower_0/mask_rcnn/mrcnn_mask/conv2d_6/Conv2D_grad/Conv2DBackpropInput}}]]
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. This isn't available when running in Eager mode.

0 successful operations.
0 derived errors ignored.

[xxxming730]

At GPU_COUNT=1 and IMAGES_PER_GPU=8, the training was completed smoothly without any memory overflow, and I felt at a loss QAQ

[CYHooo]

At GPU_COUNT=1 and IMAGES_PER_GPU=8, the training was completed smoothly without any memory overflow, and I felt at a loss QAQ

If you are training in windows,
maybe you can change strategy likes:
strategy = tf.distribute.MirroredStrategy(["GPU:0","GPU:1","GPU:2","GPU:3"],cross_device_ops=tf.distribute.ReductionToOneDevice())