Giter VIP home page Giter VIP logo

Comments (8)

Arsiuuu avatar Arsiuuu commented on June 15, 2024

因为我在MSRVTT复现得到的R1仅为44.9,训练过程中loss_ita居高不下,所以我猜想是哪些环节出了问题,希望您给予我启发!

from aurora.

xinlong-yang avatar xinlong-yang commented on June 15, 2024

loss_ita没有收敛或者变化可能是一些超参需要调整,我们之前记录的一份log日志如下,是收敛的:

{"train_lr": "0.000", "train_loss_itm": "0.270", "train_loss_ita": "6.418", "epoch": 0}
{"train_lr": "0.000", "train_loss_itm": "0.133", "train_loss_ita": "6.077", "epoch": 1}
{"train_lr": "0.000", "train_loss_itm": "0.061", "train_loss_ita": "5.849", "epoch": 2}
{"train_lr": "0.000", "train_loss_itm": "0.036", "train_loss_ita": "5.684", "epoch": 3}
{"train_lr": "0.000", "train_loss_itm": "0.028", "train_loss_ita": "5.590", "epoch": 4}
{"train_lr": "0.000", "train_loss_itm": "0.026", "train_loss_ita": "5.546", "epoch": 5}

性能gap可能是没有选择正确的.pth,我们采用base模型是指区分于blip_capfilt这种模型架构,但是预训练参数是需要根据任务调整的,调整骨架参数不影响我们对比不同baseline方法(Uniadapter, Lora)等的优劣,具体的预训练参数选择可以通过如下网站:https://storage.googleapis.com/sfr-vision-language-research

from aurora.

maoshanwen avatar maoshanwen commented on June 15, 2024

请问您复现的这个MSRVTT实验使用什么环境的呢,具体几张卡每个卡内存多少呢?

from aurora.

Arsiuuu avatar Arsiuuu commented on June 15, 2024

4*A100

from aurora.

maoshanwen avatar maoshanwen commented on June 15, 2024

那请问您数据集有改小吗?我是八卡4090,直接用readme中的语句运行,但是线程老被杀死。

from aurora.

Arsiuuu avatar Arsiuuu commented on June 15, 2024

并没有,我可以完整的运行代码,只是因为某些原因效果不太好。您能解答我上面的问题吗,那两个模块的实现在哪里?

from aurora.

maoshanwen avatar maoshanwen commented on June 15, 2024

您试着看下med.py的261看看那个信息上下文增强的,435看下那个门控的。另外能给您的pip list我看下吗?我不清楚我这个线程崩溃是内存问题还是我安装的版本不妥?(我8张4090 24G显存,搞不懂就是跑不出)

from aurora.

Arsiuuu avatar Arsiuuu commented on June 15, 2024

谢谢,435行门控是不是需要再像 https://github.com/WillDreamer/Aurora/issues/4 修改一下?

以下是我的pip list,因为是从公共环境拷贝,所以有点杂,见谅。如果您成功复现,麻烦告诉我一声。
Package Version


antlr4-python3-runtime 4.8
appdirs 1.4.4
argon2-cffi 20.1.0
astor 0.8.1
async-generator 1.10
attrs 19.3.0
autocommand 2.2.1
av 8.0.2
backcall 0.1.0
bidict 0.21.4
bleach 3.1.0
boto3 1.23.10
botocore 1.26.10
cairocffi 1.2.0
CairoSVG 2.5.2
certifi 2021.5.30
cffi 1.14.5
chardet 3.0.4
click 7.1.2
click-help-colors 0.9
cloudpickle 2.2.1
colorama 0.4.4
coloredlogs 15.0
commonmark 0.9.1
configparser 5.2.0
coverage 5.0.3
cssselect 1.1.0
cssselect2 0.4.1
cycler 0.10.0
Cython 0.29.14
dataclasses 0.8
decorator 4.4.1
decord 0.6.0
defusedxml 0.6.0
docker-pycreds 0.4.0
docutils 0.15.2
easydict 1.9
einops 0.4.1
elasticsearch 7.12.1
entrypoints 0.3
environs 9.5.0
fairscale 0.4.6
faiss 1.7.1
filelock 3.0.12
flake8 3.7.9
Flask 1.1.2
ftfy 6.0.3
future 0.18.2
gitdb 4.0.9
GitPython 3.1.18
grad-cam 1.4.8
graphviz 0.13
huggingface-hub 0.4.0
humanfriendly 9.1
humanize 3.5.0
idna 2.8
imageio 2.6.1
importlib-metadata 0.23
importlib-resources 5.4.0
iniconfig 1.1.1
ipykernel 5.1.3
ipython 7.9.0
ipython-genutils 0.2.0
ipywidgets 7.5.1
itsdangerous 1.1.0
jedi 0.15.1
Jinja2 2.10.3
jmespath 0.10.0
joblib 0.14.0
jsonschema 3.1.1
jupyter 1.0.0
jupyter-client 5.3.4
jupyter-console 6.0.0
jupyter-core 4.6.1
jupyterlab-pygments 0.1.2
kiwisolver 1.1.0
lmdb 1.4.1
lxml 4.6.3
MarkupSafe 1.1.1
marshmallow 3.14.1
matplotlib 3.1.1
mccabe 0.6.1
mistune 0.8.4
mkl-fft 1.3.0
mkl-random 1.1.1
mkl-service 2.3.0
more-itertools 7.2.0
multiprocessing-logging 0.3.4
nbclient 0.5.3
nbconvert 5.6.1
nbformat 4.4.0
nest-asyncio 1.5.1
networkx 2.4
ninja 1.10.0.post2
nltk 3.3
notebook 6.0.1
numpy 1.19.5
nvidia-dali 0.22.0
oathtool 2.3.1
olefile 0.46
omegaconf 2.1.0
onnx 1.7.0
opencv-python 4.1.1.26
packaging 21.0
pandas 1.1.5
pandocfilters 1.4.2
parso 0.5.1
path 16.2.0
pathtools 0.1.2
petrel-oss-sdk v2.2.1-2-g1505ef3-master
pexpect 4.7.0
pickleshare 0.7.5
Pillow 8.2.0
Pillow-SIMD 6.0.0.post0
pip 19.3.1
pkginfo 1.5.0.1
pluggy 0.13.1
prettytable 0.7.2
prometheus-client 0.7.1
prompt-toolkit 2.0.10
protobuf 3.19.6
psutil 5.9.6
ptyprocess 0.6.0
py 1.10.0
pybind11 2.6.2
pycocoevalcap 1.2
pycocotools 2.0.7
pycocotools-fix 2.0.0.9
pycodestyle 2.5.0
pycparser 2.20
pyDes 2.0.1
pyflakes 2.1.1
pygal 2.4.0
Pygments 2.9.0
pyparsing 3.0.9
pyrsistent 0.15.5
pytest 6.2.4
python-dateutil 2.8.1
python-dotenv 0.19.2
python-engineio 4.3.0
python-socketio 5.5.0
pytz 2021.1
PyWavelets 1.1.1
PyYAML 5.3.1
pyzmq 18.1.0
qtconsole 4.5.5
QtPy 1.9.0
readme-renderer 24.0
regex 2023.8.8
requests 2.25.1
requests-toolbelt 0.9.1
retrying 1.3.3
rich 10.1.0
ruamel.yaml 0.17.4
ruamel.yaml.clib 0.2.2
s3transfer 0.5.2
sacremoses 0.0.53
scikit-image 0.16.2
scikit-learn 0.21.3
scipy 1.3.1
seaborn 0.9.0
selenium 3.141.0
Send2Trash 1.5.0
sentry-sdk 1.34.0
setproctitle 1.2.3
setuptools 41.6.0.post20191030
six 1.15.0
smmap 5.0.0
spring 0.6.1+cu112.torch181.mvapich2.pmi2.nartgpu
spring-aux 0.6.7.develop.2021-12-29t02-21.8592b9f9
terminado 0.8.2
testpath 0.4.2
timm 0.6.12
tinycss 0.4
tinycss2 1.1.0
tokenizers 0.12.1
toml 0.10.2
torch 1.8.1+cuda112.cudnn8.1.0
torchvision 0.9.0a0+8fb5838
tornado 6.0.3
tqdm 4.37.0
traitlets 4.3.3
transformers 4.18.0
ttach 0.0.3
twine 2.0.0
typing-extensions 3.10.0.2
ujson 4.3.0
urllib3 1.26.18
wandb 0.15.11
wcwidth 0.1.7
webencodings 0.5.1
websocket-client 1.2.3
Werkzeug 1.0.1
wheel 0.33.6
widgetsnbextension 3.5.1
zipp 3.6.0

from aurora.

Related Issues (13)

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo 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.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.