Git Product home page Git Product logo

Comments (7)

okankop avatar okankop commented on September 16, 2024

You can refer to #25 for the jhmdb labels.

from yowo.

ZZZZZZZZJ avatar ZZZZZZZZJ commented on September 16, 2024

Thanks for the labels! I run the training procedure successfully.
But I have a new question that I can not reproduce the J-HMDB results.
After 25 epochs, I only get 65.07%, much lower than 74.4% shown in the paper.

AP: 98.33% (1)
AP: 98.74% (10)
AP: 97.78% (11)
AP: 39.40% (12)
AP: 40.06% (13)
AP: 94.73% (14)
AP: 71.47% (15)
AP: 30.06% (16)
AP: 31.94% (17)
AP: 14.99% (18)
AP: 45.34% (19)
AP: 22.74% (2)
AP: 71.97% (20)
AP: 64.42% (21)
AP: 91.01% (3)
AP: 76.60% (4)
AP: 94.07% (5)
AP: 57.49% (6)
AP: 56.25% (7)
AP: 69.93% (8)
AP: 99.21% (9)
mAP: 65.07%

And my jhmdb.cfg is

[net]
batch = 10
clip_duration = 16
height = 224
width = 224
channels = 3
momentum = 0.9
decay = 0.0005

learning_rate = 0.0001
max_batches = 100000
steps = 10000,20000,30000,40000
scales = 0.5,0.5,0.5,0.5


[region]
anchors = 0.95878, 3.10197, 1.67204, 4.0040, 1.75482, 5.64937, 3.09299, 5.80857$
classes = 21
num = 5

object_scale = 5
noobject_scale = 1
class_scale = 1
coord_scale = 1

my run_jhmdb-21.sh is

python train.py --dataset jhmdb-21 \
                --data_cfg cfg/jhmdb21.data \
                --cfg_file cfg/jhmdb21.cfg \
                --n_classes 21 \
                --backbone_3d resnext101 \
                --backbone_3d_weights weights/pretrained/resnext-101-kinetics-hmdb51_split1.pth \
                --freeze_backbone_3d \
                --backbone_2d darknet \
                --backbone_2d_weights weights/yolo.weights \
                --freeze_backbone_2d \

Does anything go wrong? Or if there are somethings unknown for me to influence the results?

from yowo.

okankop avatar okankop commented on September 16, 2024

Moat probably the model is overfitting. Do not wait till the end of 25 epoch to run validation. Best performance is usually achieved between 5-10 epochs.

from yowo.

kinivi avatar kinivi commented on September 16, 2024

Moat probably the model is overfitting. Do not wait till the end of 25 epoch to run validation. Best performance is usually achieved between 5-10 epochs.

Can I use negative examples for the train? For example some actions without labels (for custom dataset)?

from yowo.

west-i avatar west-i commented on September 16, 2024

@ZZZZZZZZJ
Hi,
Could you share the J-HMDB21 labels to [email protected] ?
Thanks!

from yowo.

Holmes-GU avatar Holmes-GU commented on September 16, 2024

@ZZZZZZZZJ Have you successfully reproduced the ucf24 results reported in the paper?

from yowo.

Riiick2011 avatar Riiick2011 commented on September 16, 2024

@ZZZZZZZZJ Have you successfully reproduced the ucf24 results reported in the paper?

Same question.

from yowo.

Related Issues (20)

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.