shaunyuan22 / CFINet

The official implementation for ICCV'23 paper "Small Object Detection via Coarse-to-fine Proposal Generation and Imitation Learning"
Apache License 2.0
128 stars 7 forks source link

[Feature] Pre-trained model on the SODA. #9

Closed ZachL1 closed 10 months ago

ZachL1 commented 10 months ago

What's the feature?

Great work!

Is it possible to share your pre-trained model on the SODA dataset?

Any other context?

No response

shaunyuan22 commented 10 months ago

A trained model of CFINet is available at here, and its evaluation result is as follow. Hope this could help you

+-------+-------+-------+-------+-------+-------+-------+
|AP     |AP_50  |AP_75  |AP_eS  |AP_rS  |AP_gS  |AP_N   |
+-------+-------+-------+-------+-------+-------+-------+
|30.6   |60.5   |26.7   |14.6   |27.2   |36.4   |45.1   |
+-------+-------+-------+-------+-------+-------+-------+
+---------------+-------+----------------+-------+--------------+-------+
| category      | AP    | category       | AP    | category     | AP    |
+---------------+-------+----------------+-------+--------------+-------+
| people        | 0.375 | rider          | 0.189 | bicycle      | 0.140 |
| motor         | 0.256 | vehicle        | 0.463 | traffic-sign | 0.465 |
| traffic-light | 0.388 | traffic-camera | 0.147 | warning-cone | 0.327 |
+---------------+-------+----------------+-------+--------------+-------+
ZachL1 commented 10 months ago

You are so kind, thank you for your reply! This helps me a lot!

By comparing the results of the paper, the trained model you provided should be trained on the SODA-D dataset, right?

Would you consider providing models trained on SODA-A? I ran into a problem with SODA-A, which was the same as the existing issue #8, so I posted a comment there.

Alexwong-1024 commented 10 months ago

Hello, can you use the SODA-A dataset for training on your end? @ZachL1

ZachL1 commented 10 months ago

not yet. I left it aside for now because, as you said, the SODA-A is a rotation type. This may not be what I want. I plan to try SODA-D first, and maybe I will pay attention to SODA-A when I have time in the future

Maybe @shaunyuan22 can make this clearer, such as creating a new branch in this repository focusing on SODA-A. It benefits the community and enhances the impact of this work, doesn’t it?