Evaluation Metrics for Faster-RCNN model on Own Dataset

Hello,

I am currently using the Faster_RCNN model to perform object detection on my own dataset. I am using coco_evaluation metrics and iou_threshold: 0.6, max_detection_per_class: 100, max_total_detections:300 parameter values in the Faster_RCNN model config file.

When evaluating the trained Faster_RCNN model on my dataset, I am getting a 0.55 mAP for large objects. However, when I perform a prediction directly on images, it looks better than the actual mAP score I am getting from the evaluation. I suspect the evaluation metrics and parameters might be required to change to fit my problem. Does anyone have similar experience dealing with this problem? If I need to change evaluation metrics or parameters in the config file, what changes can be made to get a better mAP?

Thank you so much for your help!

I have written my complete config file below.

model {
faster_rcnn {
num_classes: 6
image_resizer {
keep_aspect_ratio_resizer {
min_dimension: 600
max_dimension: 1024
}
}
feature_extractor {
type: 'faster_rcnn_inception_v2'
first_stage_features_stride: 16
}
first_stage_anchor_generator {
grid_anchor_generator {
scales: [0.25, 0.5, 1.0, 2.0]
aspect_ratios: [0.5, 1.0, 2.0]
height_stride: 16
width_stride: 16
}
}
first_stage_box_predictor_conv_hyperparams {
op: CONV
regularizer {
l2_regularizer {
weight: 0.0
}
}
initializer {
truncated_normal_initializer {
stddev: 0.01
}
}
}
first_stage_nms_score_threshold: 0.0
first_stage_nms_iou_threshold: 0.7
first_stage_max_proposals: 300
first_stage_localization_loss_weight: 2.0
first_stage_objectness_loss_weight: 1.0
initial_crop_size: 14
maxpool_kernel_size: 2
maxpool_stride: 2
second_stage_box_predictor {
mask_rcnn_box_predictor {
use_dropout: false
dropout_keep_probability: 1.0
fc_hyperparams {
op: FC
regularizer {
l2_regularizer {
weight: 0.0
}
}
initializer {
variance_scaling_initializer {
factor: 1.0
uniform: true
mode: FAN_AVG
}
}
}
}
}
second_stage_post_processing {
batch_non_max_suppression {
score_threshold: 0.0
iou_threshold: 0.6
max_detections_per_class: 100
max_total_detections: 300
}
score_converter: SOFTMAX
}
second_stage_localization_loss_weight: 2.0
second_stage_classification_loss_weight: 1.0
}
}

train_config: {
batch_size: 1
optimizer {
momentum_optimizer: {
learning_rate: {
manual_step_learning_rate {
initial_learning_rate: 0.0002
schedule {
step: 900000
learning_rate: .00002
}
schedule {
step: 1200000
learning_rate: .000002
}
}
}
momentum_optimizer_value: 0.9
}
use_moving_average: false
}
gradient_clipping_by_norm: 10.0
fine_tune_checkpoint: "/home/models/research/object_detection/faster_rcnn_inception_v2_coco_2018_01_28/model.ckpt"
from_detection_checkpoint: true
load_all_detection_checkpoint_vars: true

num_steps: 200000
data_augmentation_options {
random_horizontal_flip {
}
}
}

train_input_reader: {
tf_record_input_reader {
input_path: "/data/datasets/mydataset/train_clear_day/train_clear_day_*.swedentfrecord"
}
label_map_path: "/home/models/research/object_detection/data/mydata_label_map.pbtxt"
}

eval_config: {
metrics_set: "coco_detection_metrics"
num_examples: 200
max_evals: 10
}

eval_input_reader: {
tf_record_input_reader {
input_path: "/data/datasets/mydataset/val_clear_day/val_clear_day_*.swedentfrecord"
}
label_map_path: "/home/models/research/object_detection/data/mydata_label_map.pbtxt"
shuffle: false
num_readers: 1
}