Comments (3)
Hi,
When you generalize to multi-class, you should first consider if you want the IoU head (as well as the standard regression head) to be class agnostic. In our experiments it worked well.
In your experiment, you have 40 different IoU scores per detection, one for each class. You can try generalizing the inference stage for class-dependent IoU score in various ways, but I recommend that you first try training a single IoU head for all classes.
As for the network you already trained, there are several possible experiments that can be made. Simple approached to reduce the 40 scores into 1 are taking the average, median, maximum, etc.
For a more elegant approaches you may consider some version of normalized weighted average of the 40 numbers, weighed according to their classification scores.
Regards,
Eran
from sku110k_cvpr19.
In addition, if you encounter problems with the multi-class version of the classification head, I think it would be best to look at the original code of keras-retinanet as a reference.
from sku110k_cvpr19.
Can anyone show me an example of how to set up a class-agnostic IoU and regression head?
from sku110k_cvpr19.
Related Issues (20)
- Wrong classes in dataset, just "object" HOT 3
- Fine-tuning model on custom dataset, num classes > 1
- Where to find annotations_test.csv file? HOT 1
- how to change default directory HOT 1
- Suggestions for compatibility HOT 7
- About the phenomenon of nan in training
- Module 'keras.backend' has no atribute 'tensorflow_backend' HOT 3
- No module named 'keras_resnet' HOT 2
- Please check the score (Table 4) in your paper. HOT 4
- found a bug HOT 1
- Request for sku110k pretrained base model for fine-tuning with own dataset HOT 2
- Is counting result available? Also can this code be used to for a custom dataset or is it just for the SKU110k dataset ?
- Em Merger merge_detections return nothing
- ValueError: invalid CSV annotations file annotations_train.csv: line 1494: y2 (799) must be higher than y1 (799) HOT 1
- Incorrect results - what am I doing wrong? HOT 1
- ValueError: signal only works in main thread
- In predict.py CLA --iou-threshold is not used
- Does anyone implement it with pytorch?
- 110K Category Labels: Each object is labelled as 'object', can you please share the labels of each object? HOT 1
- Training problems in YOLOv5 HOT 5
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
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.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from sku110k_cvpr19.