Improve the model performance of SegNet

Hi,

I am currently using the same Mask R-CNN model in the SegNet for fish tracking, and the results seem to be not enough satisfying (see below). I am a new-comer for deep learning projects, so any suggestions from you to improve my model (reduce the losses or accelerate the tracking speed) is always welcomed. trout20220724T2116_sum My general setups are:

epoch for head layer = 5;
epoch for all layers = 60;
training datasets = 313 images;
validation datasets = 123 images;

Note that all images in the datasets have been pre-processed using the backgroundsubtractorKNN. A sample image is displayed as follow (white areas are fishes that I need to track, 3 in maximum):

I also list my configurations below:

 Configurations: 

BACKBONE                       resnet50
BACKBONE_STRIDES               [4, 8, 16, 32, 64]
BATCH_SIZE                     6
BBOX_STD_DEV                   [0.1 0.1 0.2 0.2]
COMPUTE_BACKBONE_SHAPE         None
DETECTION_MAX_INSTANCES        35
DETECTION_MIN_CONFIDENCE       0.8
DETECTION_NMS_THRESHOLD        0.3
FPN_CLASSIF_FC_LAYERS_SIZE     1024
GPU_COUNT                      1
GRADIENT_CLIP_NORM             5.0
IMAGES_PER_GPU                 3
IMAGE_CHANNEL_COUNT            3
IMAGE_MAX_DIM                  1024
IMAGE_META_SIZE                14
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': 0.5, 'mrcnn_mask_loss': 0.5}
MASK_POOL_SIZE                 14
MASK_SHAPE                     [28, 28]
MAX_GT_INSTANCES               5
MEAN_PIXEL                     [123.7 116.8 103.9]
MINI_MASK_SHAPE                (56, 56)
NAME                           trout
NUM_CLASSES                    2
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              (16, 32, 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                200
TOP_DOWN_PYRAMID_SIZE          256
TRAIN_BN                       False
TRAIN_ROIS_PER_IMAGE           250
USE_MINI_MASK                  False
USE_RPN_ROIS                   True
VALIDATION_STEPS               50
WEIGHT_DECAY                   0.0001

Thanks in advance for taking your time to check my problems. Any help is appreciated ！

Best, Erin

SIPEC-Animal-Data-Analysis / SIPEC

Improve the model performance of SegNet #32