-
Say my dataset consists of images with cats and dogs. Some images may have both the dog and the cat but the label for the cat would be missing.
Is there a straightforward way and fast way to modif…
-
I have a question about CPL, in the code:models/flexmatch/flexmatch_utils.py line 44~46.
```
44 mask = max_probs.ge(p_cutoff * (class_acc[max_idx] / (2. - class_acc[max_idx]))).float() # convex
…
-
One link you post in your MoA_competition/Models/MoA_TabNet.ipynb, which is following
test-moa-public: https://www.kaggle.com/nicohrubec/test-public-moa now is missing.
Can you send me that "test-pu…
-
現在のLBはCVに対してかなり高い。
特に公開コードはLB overfitに見えるのでデータ傾向がpublic/privateで異なる場合に大きくshake down する可能性がある。
[考えたいデータ傾向]
- 分類の難易度(今出ている公開コードでは左右差がある系のターゲットと予測がうまくいっていない)
- GBDT stackingした方がCVでは改善するがLBは悪化
- …
-
Thanks for the great work on debiasing pseudo-labeling!
Regarding Eqn(5), it is mentioned in the paper that [R(1)] adaptive margin loss can demand a larger margin between hardly biased and highly b…
-
When you train your CRAFT model in a weakly-supervised manner:
- What are your transformations in transforms.Compose, which transformation and in which order?
- Did you move your batch of preprocess…
-
### Backbone
* EfficientNet
* https://github.com/zhanghang1989/ResNeSt
### Pre-trained model
* better pre-trained model than ImageNet-1.2M pre-trained model
- Weakly supervised Instagram-1G (…
-
This is an RFC for the implementation of a multithreaded version of the connected component labeling algorithm.
Such algorithms have appeared in research papers since a few years ago. A common theme…
-
Hi,
Thanks for the interesting work and sharing the code.
Recently, I reproduced the Fine-Tuning baseline method based on the released code for Self-Tuning method (directly delete the unlabeled …
-