TL Summary

exp id	model id	optimizer	img size	batch size	lr head	lr ft	dropout	finetune	unfreeze pct	head epochs	ft epochs	accuracy	macro f1	weighted f1	best val acc	classes	samples	imbalance ratio
best	MNv3l	adamw	224	64	0.001	7e-05	0.4	True	0.5	12	35	0.9964	0.9964	0.9964	0.9933	26	15921	1.0300
exp_tl_001	MNv3l	adamw	224	64	0.001	7e-05	0.4	True	0.5	12	35	0.9964	0.9964	0.9964	0.9933	26	15921	1.0300
exp_tl_004	ResNet50	adamw	224	48	0.001	0.0001	0.4	True	0.3	10	30	0.9959	0.9959	0.9959	0.9928	26	15921	1.0300
exp_tl_002	EffB0	adamw	224	64	0.001	5e-05	0.4	True	0.3	10	30	0.9949	0.9949	0.9949	0.9938	26	15921	1.0300
exp_tl_003	MNv2	adamw	224	64	0.001	0.0001	0.4	True	0.35	12	30	0.9933	0.9933	0.9933	0.9938	26	15921	1.0300
exp_tl_005	Xception	adamw	224	48	0.001	0.0001	0.4	True	0.35	10	30	0.9928	0.9928	0.9928	0.9949	26	15921	1.0300
exp_tl_006	InceptionV3	adamw	299	32	0.001	0.0001	0.4	True	0.3	10	25	0.9897	0.9897	0.9897	0.9892	26	15921	1.0300
exp_tl_008	VGG16	adamw	224	32	0.001	1e-05	0.5	True	0.25	8	20	0.9877	0.9877	0.9877	0.9882	26	15921	1.0300
exp_tl_007	MNv3s	adamw	224	96	0.001	0.0001	0.3	True	0.4	12	30	0.9867	0.9867	0.9867	0.9867	26	15921	1.0300

best

Acc 0.9964 MacroF1 0.9964 WeightedF1 0.9964 Best Val 0.9933

EDA: kelas: 26 · sampel: 15921 · min/max per kelas: 601/619 · imbalance: 1.0300

README: ada

Config

{
  "model_id": "MNv3l",
  "img_size": 224,
  "batch_size": 64,
  "optimizer": "adamw",
  "lr_head": 0.001,
  "lr_ft": 7e-05,
  "weight_decay": 1e-05,
  "label_smoothing": 0.05,
  "dropout": 0.4,
  "head_epochs": 12,
  "ft_epochs": 35,
  "finetune": true,
  "unfreeze_pct": 0.5,
  "lr_schedule": "reduce",
  "patience": 9
}

Plots

Artifacts

keras — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\best\artifacts\best_ft.keras
h5 — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\best\artifacts\final_model.h5
tflite — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\best\artifacts\model_fp16.tflite

exp_tl_001

Acc 0.9964 MacroF1 0.9964 WeightedF1 0.9964 Best Val 0.9933

EDA: kelas: 26 · sampel: 15921 · min/max per kelas: 601/619 · imbalance: 1.0300

README: ada

Config

{
  "model_id": "MNv3l",
  "img_size": 224,
  "batch_size": 64,
  "optimizer": "adamw",
  "lr_head": 0.001,
  "lr_ft": 7e-05,
  "weight_decay": 1e-05,
  "label_smoothing": 0.05,
  "dropout": 0.4,
  "head_epochs": 12,
  "ft_epochs": 35,
  "finetune": true,
  "unfreeze_pct": 0.5,
  "lr_schedule": "reduce",
  "patience": 9
}

Plots

Artifacts

keras — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_001\artifacts\best_ft.keras
h5 — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_001\artifacts\final_model.h5
tflite — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_001\artifacts\model_fp16.tflite

exp_tl_002

Acc 0.9949 MacroF1 0.9949 WeightedF1 0.9949 Best Val 0.9938

EDA: kelas: 26 · sampel: 15921 · min/max per kelas: 601/619 · imbalance: 1.0300

README: ada

Config

{
  "model_id": "EffB0",
  "img_size": 224,
  "batch_size": 64,
  "optimizer": "adamw",
  "lr_head": 0.001,
  "lr_ft": 5e-05,
  "weight_decay": 1e-05,
  "label_smoothing": 0.05,
  "dropout": 0.4,
  "head_epochs": 10,
  "ft_epochs": 30,
  "finetune": true,
  "unfreeze_pct": 0.3,
  "lr_schedule": "reduce",
  "patience": 8
}

Plots

Artifacts

keras — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_002\artifacts\best_ft.keras
h5 — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_002\artifacts\final_model.h5
tflite — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_002\artifacts\model_fp16.tflite

exp_tl_003

Acc 0.9933 MacroF1 0.9933 WeightedF1 0.9933 Best Val 0.9938

EDA: kelas: 26 · sampel: 15921 · min/max per kelas: 601/619 · imbalance: 1.0300

README: ada

Config

{
  "model_id": "MNv2",
  "img_size": 224,
  "batch_size": 64,
  "optimizer": "adamw",
  "lr_head": 0.001,
  "lr_ft": 0.0001,
  "weight_decay": 1e-05,
  "label_smoothing": 0.05,
  "dropout": 0.4,
  "head_epochs": 12,
  "ft_epochs": 30,
  "finetune": true,
  "unfreeze_pct": 0.35,
  "lr_schedule": "reduce",
  "patience": 8
}

Plots

Artifacts

keras — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_003\artifacts\best_ft.keras
h5 — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_003\artifacts\final_model.h5
tflite — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_003\artifacts\model_fp16.tflite

exp_tl_004

Acc 0.9959 MacroF1 0.9959 WeightedF1 0.9959 Best Val 0.9928

EDA: kelas: 26 · sampel: 15921 · min/max per kelas: 601/619 · imbalance: 1.0300

README: ada

Config

{
  "model_id": "ResNet50",
  "img_size": 224,
  "batch_size": 48,
  "optimizer": "adamw",
  "lr_head": 0.001,
  "lr_ft": 0.0001,
  "weight_decay": 1e-05,
  "label_smoothing": 0.05,
  "dropout": 0.4,
  "head_epochs": 10,
  "ft_epochs": 30,
  "finetune": true,
  "unfreeze_pct": 0.3,
  "lr_schedule": "reduce",
  "patience": 8
}

Plots

Artifacts

keras — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_004\artifacts\best_ft.keras
h5 — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_004\artifacts\final_model.h5
tflite — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_004\artifacts\model_fp16.tflite

exp_tl_005

Acc 0.9928 MacroF1 0.9928 WeightedF1 0.9928 Best Val 0.9949

EDA: kelas: 26 · sampel: 15921 · min/max per kelas: 601/619 · imbalance: 1.0300

README: ada

Config

{
  "model_id": "Xception",
  "img_size": 224,
  "batch_size": 48,
  "optimizer": "adamw",
  "lr_head": 0.001,
  "lr_ft": 0.0001,
  "weight_decay": 1e-05,
  "label_smoothing": 0.05,
  "dropout": 0.4,
  "head_epochs": 10,
  "ft_epochs": 30,
  "finetune": true,
  "unfreeze_pct": 0.35,
  "lr_schedule": "reduce",
  "patience": 8
}

Plots

Artifacts

keras — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_005\artifacts\best_ft.keras
h5 — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_005\artifacts\final_model.h5
tflite — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_005\artifacts\model_fp16.tflite

exp_tl_006

Acc 0.9897 MacroF1 0.9897 WeightedF1 0.9897 Best Val 0.9892

EDA: kelas: 26 · sampel: 15921 · min/max per kelas: 601/619 · imbalance: 1.0300

README: ada

Config

{
  "model_id": "InceptionV3",
  "img_size": 299,
  "batch_size": 32,
  "optimizer": "adamw",
  "lr_head": 0.001,
  "lr_ft": 0.0001,
  "weight_decay": 1e-05,
  "label_smoothing": 0.05,
  "dropout": 0.4,
  "head_epochs": 10,
  "ft_epochs": 25,
  "finetune": true,
  "unfreeze_pct": 0.3,
  "lr_schedule": "reduce",
  "patience": 8
}

Plots

Artifacts

keras — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_006\artifacts\best_ft.keras
h5 — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_006\artifacts\final_model.h5
tflite — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_006\artifacts\model_fp16.tflite

exp_tl_007

Acc 0.9867 MacroF1 0.9867 WeightedF1 0.9867 Best Val 0.9867

EDA: kelas: 26 · sampel: 15921 · min/max per kelas: 601/619 · imbalance: 1.0300

README: ada

Config

{
  "model_id": "MNv3s",
  "img_size": 224,
  "batch_size": 96,
  "optimizer": "adamw",
  "lr_head": 0.001,
  "lr_ft": 0.0001,
  "weight_decay": 1e-05,
  "label_smoothing": 0.05,
  "dropout": 0.3,
  "head_epochs": 12,
  "ft_epochs": 30,
  "finetune": true,
  "unfreeze_pct": 0.4,
  "lr_schedule": "reduce",
  "patience": 8
}

Plots

Artifacts

keras — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_007\artifacts\best_ft.keras
h5 — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_007\artifacts\final_model.h5
tflite — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_007\artifacts\model_fp16.tflite

exp_tl_008

Acc 0.9877 MacroF1 0.9877 WeightedF1 0.9877 Best Val 0.9882

EDA: kelas: 26 · sampel: 15921 · min/max per kelas: 601/619 · imbalance: 1.0300

README: ada

Config

{
  "model_id": "VGG16",
  "img_size": 224,
  "batch_size": 32,
  "optimizer": "adamw",
  "lr_head": 0.001,
  "lr_ft": 1e-05,
  "weight_decay": 1e-05,
  "label_smoothing": 0.0,
  "dropout": 0.5,
  "head_epochs": 8,
  "ft_epochs": 20,
  "finetune": true,
  "unfreeze_pct": 0.25,
  "lr_schedule": "reduce",
  "patience": 6
}

Plots

Artifacts

keras — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_008\artifacts\best_ft.keras
h5 — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_008\artifacts\final_model.h5
tflite — D:\Dhika\(KULIAH)\'Skripsi\Github\Fix_TA_Project\Fix_TA_Project\1. Workspace\3. Transferlearning\runs\exp_tl_008\artifacts\model_fp16.tflite

Leaderboard Transfer Learning

best

Plots

Artifacts

exp_tl_001

Plots

Artifacts

exp_tl_002

Plots

Artifacts

exp_tl_003

Plots

Artifacts

exp_tl_004

Plots

Artifacts

exp_tl_005

Plots

Artifacts

exp_tl_006

Plots

Artifacts

exp_tl_007

Plots

Artifacts

exp_tl_008

Plots

Artifacts