zhengli97 / ctkd Goto Github PK
View Code? Open in Web Editor NEW[AAAI 2023] Official PyTorch Code for "Curriculum Temperature for Knowledge Distillation"
Home Page: https://zhengli97.github.io/CTKD/
License: Apache License 2.0
[AAAI 2023] Official PyTorch Code for "Curriculum Temperature for Knowledge Distillation"
Home Page: https://zhengli97.github.io/CTKD/
License: Apache License 2.0
您好,请问num_loops为什么设为10,以及10个epoch之后lambda的值就不变了吗,我看code是这样写的,若不变了是怎么加大课程的难度的呢?
请删除 models/__init__.py
里面的第 14 行
from .temp_mlp import Temp_MLP1
然后如果方便的话,请提供一个Requirements.txt
您好!阅读了你们的CTKD,对这工作十分感兴趣。我在yolov8上做蒸馏使用CTKD,开始时训练temp不会改变,发现可能是使用了amp导致的,关amp闭后temp一直增大,发现并不会经过GRL的backward。请问您有什么见解,谢谢谢谢。
Thanks for your great work.
We try to reproduce reults of KD (without CTKD) using run_imagenet_distill.sh
but t it seems that something is wrong.
python3 train_student.py --path-t ./save/models/ResNet34/resnet34.pth \
--distill kd \
--batch_size 256 --epochs 120 --dataset imagenet \
--gpu_id 0,1,2,3,4,5,6,7 --num_workers 32 \
--learning_rate 0.1 --lr_decay_epochs 30,60,90 --weight_decay 1e-4 \
--model_s ResNet18 -r 1 -a 1 -b 0 --kd_T 4 \
...
RN34 | RN18 | KD | |
---|---|---|---|
Paper | 73.96 | 70.26 | 70.83 |
Ours | 73.79 | 70.34 | 70.27, 70.48 |
I would like to know whether the cause is experimental error or parameter error.
你好,我是将你们code里跟temp有关的部分替换了我自己的蒸馏方法里的温度,在第一次loss.backward后,temp的值从15.6212变成了nan,且一直是nan,请问有什么问题或建议吗
您好,请问一下,instance-T算法在代码的何处呢?我没找到。
您好,请问交替优化机制和GRL梯度翻转层在代码里的哪处呢?
我在您提供的代码中只看到了global-T的实现,并没有找到instance-τ具体是怎么做的。可能是您忘了上传吗?或者是我没有找到?请问可以提供一下吗?
Why is there no code for CTKD in the run_cifar_distill.sh file? And the code about 14 lines of DKD seems to be wrong!
请问一下,为什么代码实现KD时,教师模型的输出是经过log_softmax,而学生模型输出经过softmax,会有什么影响吗(除了log_softmax计算快)
KD_loss += nn.KLDivLoss(reduction='batchmean')(F.log_softmax(y_s/T, dim=1), F.softmax(y_t/T, dim=1)) * T * T
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.