Inconsistent result of coco metrics and eval side-by-side.

[3zhang]

Prerequisites

Please answer the following questions for yourself before submitting an issue.

• I am using the latest TensorFlow Model Garden release and TensorFlow 2.
• I am reporting the issue to the correct repository. (Model Garden official or research directory)
• I checked to make sure that this issue has not already been filed.

1. The entire URL of the file you are using

https://www.kaggle.com/datasets/marcosgabriel/photovoltaic-system-thermography

2. Describe the bug

I trained a mask-rcnn model (fine-tuned from mask_rcnn_inception_resnet_v2_1024x1024_coco17_gpu-8) based on the dataset above. The training was done successfully. However the evaluation result is confusing. The AP@IoU=0.50/0.75 are very high (~1). But based on the side-by-side image result, it should not be that high.

Evaluate annotation type bbox
DONE (t=0.14s).
Accumulating evaluation results...
DONE (t=0.01s).
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.815
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.999
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.999
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = -1.000
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.815
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = -1.000
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.040
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.399
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.854
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = -1.000
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.854
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = -1.000
creating index...
index created!
creating index...
index created!
Running per image evaluation...
Evaluate annotation type segm
DONE (t=0.14s).
Accumulating evaluation results...
DONE (t=0.01s).
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.836
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.999
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.999
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = -1.000
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.836
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = -1.000
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.042
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.409
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.874
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = -1.000
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.874
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = -1.000z

3. Steps to reproduce

eval config:
eval_config: {
metrics_set: "coco_detection_metrics"
metrics_set: "coco_mask_metrics"
eval_instance_masks: true
use_moving_averages: false
batch_size: 2
include_metrics_per_category: false
}

eval_input_reader: {
label_map_path: "/content/labelmap.txt"
shuffle: false
num_epochs: 1
tf_record_input_reader {
input_path: "/content/tf_record/sample_val.record"
}
load_instance_masks: true
mask_type: PNG_MASKS
}

4. Expected behavior

By roughly manually calculation, the precision is 0.99382716049, the recall is 0.92528735632. I know this is not AP. But it clearly shows that AP should not be ~1.

5. Additional context

6. System information

Evaluation was done on google colab.