Comments (7)
Of course! cuz the step of the save best
is 2600, I guess that you have used the pre-trained model. Can you show more details of your config of the training, like example.json
? @BUAAtao
from chexpert.
I didn't change other parameters except for train.csv's path. I also trained the model by 4 1080ti.
{
"train_csv": "/data/Chexpert/CheXpert-v1.0/train.csv",
"dev_csv": "/data/Chexpert/CheXpert-v1.0/valid.csv",
"backbone": "densenet121",
"width": 512,
"height": 512,
"long_side": 512,
"fix_ratio": true,
"pixel_mean": 128.0,
"pixel_std": 64.0,
"use_pixel_std": true,
"use_equalizeHist": true,
"use_transforms_type": "Aug",
"gaussian_blur": 3,
"border_pad": "pixel_mean",
"num_classes": [1,1,1,1,1],
"batch_weight": true,
"enhance_index": [2,6],
"enhance_times": 2,
"pos_weight": [1,1,1,1,1],
"train_batch_size": 56,
"dev_batch_size": 56,
"pretrained": true,
"log_every": 10,
"test_every": 100,
"epoch": 3,
"norm_type": "BatchNorm",
"global_pool": "AVG_MAX",
"fc_bn": true,
"attention_map": "FPA",
"lse_gamma": 0.5,
"fc_drop": 0,
"optimizer": "Adam",
"criterion": "BCE",
"lr": 0.0001,
"lr_factor": 0.1,
"lr_epochs": [2],
"momentum": 0.9,
"weight_decay": 0.0,
"best_target": "auc",
"save_top_k": 3,
"save_index": [0,1,2,3,4]
}
from chexpert.
And I didn't use the pre_trained .pth, why the step is 2600 is that I used the best performance model (best3.ckpt) to test. And the best1.ckpt's step is 2000.
from chexpert.
fine.In my opinion,
- larger batch size will get better results(I have tried).
- the
save_index
is the importance of saving models for different diseases. (that means that Your preservation strategy focuses more on which diseases), the Cardiomegaly and Atelectasis are more unstable.
3.try the "enhance_index": [2,6],
"enhance_times": 1, in the config file.(that means oversampling) @BUAAtao
from chexpert.
Thanks. Based on your advice, I have changed "enhance_times" to 1. And get a better result as follow:
Save best is step : 8000 AUC : 0.8934086684552691
Cardiomegaly auc 0.8524423337856174
Edema auc 0.9329415310427969
Consolidation auc 0.9440104166666667
Atelectasis auc 0.84816
Pleural_Effusion auc 0.9105009191176471
I couldn't change the batch size because of the nums of GPU. I also found that he Cardiomegaly and Atelectasis were unstable and don't know how to do some changes to "save_index".Could you tell me more details about "save_index"? Then I would try a few more times to get a better restult.
from chexpert.
you can find the "save_index" is [0,1,2,3,4] in the config/example
.That means the best model you saved is the higher mean of five diseases. If you set it as [0,3], it will save the model According to the average of the two results. (Cardiomegaly and Atelectasis) @BUAAtao
from chexpert.
Thanks.I changed “save_index” to [0,3] and trained the model again.The result is better. But the result is still not great.I would try other ways to improve my model.Thanks for your advice again!
from chexpert.
Related Issues (20)
- Definiton of PcamPool in global_pool.py HOT 2
- Questions about normalization and auc score HOT 6
- Use pre-trained model on customised dataset HOT 1
- use equalizeHist enhance the images before feed images into network? HOT 3
- Reproducible Result Problem HOT 4
- Can you add a license file? HOT 3
- 请问提供的预训练模型是AVG_MAX的吗? HOT 2
- Lower AUC score of pre-trained weights compared to the README HOT 2
- Model added to the torchxrayvision library HOT 1
- Example Configuration HOT 9
- Tensorflow trained model HOT 4
- Are you sincere? HOT 3
- question about the tensor size HOT 1
- Output probabilities of all 14 classes HOT 2
- Predict an image from path
- About changing dataset
- On the uncertain label problem
- RuntimeError: stack expects a non-empty TensorList HOT 1
- What's the pretrained model?
- Discrepancy in AUC Scores Using Provided Pre-trained Model
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from chexpert.