DetectionTeamUCAS / DetectionTeamUCAS/RetinaNet_Tensorflow_Rotation
FPN中RoI Pooling层的公式疑惑
- Dominant language
- Python
- Stars
- 312
- Forks
- 92
- PR merge metrics
- No merged PRs in 30d
Description
老哥老哥,还有一个疑问,自己想了半天也没明白,网上也找不到:
FPN网络里在RoI Pooling层的部分是将不同的RoI分配给金字塔的不同层,使用的分配公式是:`levels = tf.round(4. + tf.log(tf.sqrt(w*h + 1e-8)/224.0) / tf.log(2.))`这里的224是标准ImageNet训练前的图片大小。
所以我的疑问是咱们在训练自己的数据集时,假如我的输入图片大小是800×800,为啥在RoI Pooling的这个公式里还是用的224而不是800呢?这个是跟ResNets用ImageNet预训练出来的有关系吗?
有点懵,望老哥指点一二,抱拳了~
Contributor guide
No contributing guide indexed for this repository
Research direction
Start with the RoI Pooling level-assignment formula quoted in the issue and locate its FPN implementation in the repository. Trace how w, h, the 224 constant, and the input image scale are defined, then document the relationship between this formula and the ResNet/ImageNet setup. Done means the repository explains why 224 is used and how the rule applies to custom image sizes.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, tensorflow
- Domain
- computer-vision, machine-learning
- Issue type
- Documentation
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100