Rethinking Data-augmentation

Fixing Resolution[1]

Image Super-Resolution[2] 적용?
Super-Resolution는 이미지 size 2x 4x 올려가는데 미세 정봈까지 정확하게 나올 수 있는 CV 연구입니다.

3D GAN[3] 걱용?

Inpaint[4] 적용?

Rethinking EfficientNet[5]

Compound Model Scaling: depth，width，resolution ($\phi$) 어떤 depth，width，resolution 파라메타가 좋고 빠를까 $\phi$ 통해서 바꿀 수 있다.

사용한 factorized hierarchical search space 방법은 Mnasnet[6] 같다. EfficientNet v2[7]
Progressive Learning: Training 과정 중 이미지의 크기 및 model 강한 정규화를 적용하아 하는것을 제안한다.

ex-work

yolo-v5 사용해서 작은 이미지 dataset 만들었다.

Next work

1.Open Source Data Labeling/Annotation 정리하기
2.Data-augmentation 새방식 + EfficientNet 새로운 연구 조사 및 coding 실험
3.실험 결과 분석 및 보고서 작성

References

[1] Touvron H, Vedaldi A, Douze M, et al. Fixing the train-test resolution discrepancy[J]. Advances in neural information processing systems, 2019, 32.
[2] Wang X, Xie L, Dong C, et al. Real-esrgan: Training real-world blind super-resolution with pure synthetic data[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. 2021: 1905-1914.
[3] Chan E R, Lin C Z, Chan M A, et al. Efficient geometry-aware 3D generative adversarial networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2022: 16123-16133.
[4] Lugmayr A, Danelljan M, Romero A, et al. Repaint: Inpainting using denoising diffusion probabilistic models[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2022: 11461-11471.
[5] Tan M, Le Q. Efficientnet: Rethinking model scaling for convolutional neural networks[C]//International conference on machine learning. PMLR, 2019: 6105-6114.
[6] Tan M, Chen B, Pang R, et al. Mnasnet: Platform-aware neural architecture search for mobile[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2019: 2820-2828.
[7] Tan M, Le Q. Efficientnetv2: Smaller models and faster training[C]//International Conference on Machine Learning. PMLR, 2021: 10096-10106.