competition_iqiyi

Ref

• mtcnn
• insightface
• InsightFace_TF

Dataset Details

Training Set Statistical Information (in log scale)

Validation Set Statistical Information (in log scale)

Experiment

Overview Report

method	TRAIN mAP	VAL mAP	TRAIN Precision (infer_2)	VAL Precision	Test mAP
mtcnn + insightface + means knn (baseline)	none	0.771192	none	0.82217	none
mtcnn + insightface + our FC + means knn	none	0.365657	0.89025	0.4204	none

Partial Training Report

method	TRAIN Loss	TRAIN Precision (infer)	VAL Precision
our FC (512in_2048_512out)	2.01	0.904674	0.365657

InsightFace+FC Usage:

1. Git clone the InsightFace_TF reposity.
2. Train the model by train_fc_only.py script in InsightFace repo.
3. Evaluate the validate set by eval_map_fc_only.py script.

Error Report Usage:

1. Run my report.py
2. Then we will get a report.md
3. export report.pdf by using the largest report.md printing page size.