# 2022 / 10 / 26 進度 ###### tags: `實驗` [TOC] ### 資料 Train / Valid 8 : 2 | - | Train | Valid | | -------- | -------- | -------- | | Benign | 225 | 56 | | Normal | 80 | 20 | | Malignant | 56 | 18 | ### 做了分Benign+Normal / Malignant的實驗 | Model | Best AUC | Best AUC - Acc | Threshold | | -------- | -------- | -------- | ------- | | Unblance Data | 0.8735 | 0.8298 | | Blance Data | 0.9211 | 0.8723 | | Add Last One Feature Mix | 0.9247 | <font color=red>0.9468 | - | | Add Last One Feature Mix | 0.9408 | <font color=red>0.9043 | 0.67385495 | | Add Last One Predict | <font color=red>0.9327 | 0.9255 | ### --------------------------------------------- | Model | Best Acc | Best Acc - AUC | | -------- | -------- | -------- | | Unblance Data | 0.9255 | 0.8480 | | Blance Data | 0.9255 | 0.8904 | | Add Last One Feature Mix | <font color=red>0.9468</font> | <font color=red>0.9247</font> | | Add Last One Predict | 0.9362 | 0.9181 | 切Patch的說法 挑3~10張 Normal/Benign/Malignant | Model | Best AUC | | -------- | -------- | | Baseline | 0.8735 | | Random balance training data | 0.9211 | | Add previous bone-scan in Linear | 0.9327 | | Add previous bone-scan features mix | <font color=red>0.9408 | | Patch-base model | mAP50 | mAR50:95 | | -------- | -------- | -------- | | Baseline | 0.5422 | 0.5982 | | Remove the cut out boxes | <font color=red>0.6879 | 0.6223 | | Overlap patches | 0.6304 | <font color=red>0.6257 | | Model | mAP50 | mAR50:95 | Time (epoch = 100) | | -------- | -------- | -------- | -------- | | Faster-RCNN | 0.2254 | 0.2750 | 75min | | FCOS | <font color=red>0.2456 | <font color=red>0.2804 | <font color=red>65min | | YOLO v7 | 0.2173 | - | 100min | | Augmentation | mAP50 | mAR50:95 | | -------- | -------- | -------- | | Baseline | 0.5422 | 0.5982 | | Brightness | 0.5495 | 0.5976 | | Horizontal Flip | <font color=red>0.5619 | 0.6205 | | Rotate in 5 | 0.5599 | <font color=red>0.6279 |
×
Sign in
Email
Password
Forgot password
or
By clicking below, you agree to our
terms of service
.
Sign in via Facebook
Sign in via Twitter
Sign in via GitHub
Sign in via Dropbox
Sign in with Wallet
Wallet (
)
Connect another wallet
New to HackMD?
Sign up