Comments (7)
Have you cheacked your training data? Your training captions may contain extra spaces which leads to this.
from clip_prefix_caption.
there is not space in train data.And when inference, it's just need the clip_embedding tensor,not need tha caption.isn't it?
from clip_prefix_caption.
I find that not all of your training caption end with the '.', since the end token for beamsearch is '.', thus the model may not know when to end inference but keeps predcting the 'space' till reach the max inference length. Yes, there's no need to input the caption when inference, i mean the training data will influence your model's ability when inference. Maybe you can process the training caption to all end with '.' and re-training your model to have a try?
from clip_prefix_caption.
I find that not all of your training caption end with the '.', since the end token for beamsearch is '.', thus the model may not know when to end inference but keeps predcting the 'space' till reach the max inference length. Yes, there's no need to input the caption when inference, i mean the training data will influence your model's ability when inference. Maybe you can process the training caption to all end with '.' and re-training your model to have a try?
thank you. I will try it.
from clip_prefix_caption.
I got the same situation, may I ask that if this problem is solved through "process the training caption to all end with '.'" ?
from clip_prefix_caption.
I really want to know how you generate this caption, I am new to this direction, do not know how you run the result, I follow the readme run, it seems that I can not evaluate the model and predict the caption
from clip_prefix_caption.
I really want to know how you generate this caption, I am new to this direction, do not know how you run the result, I follow the readme run, it seems that I can not evaluate the model and predict the caption
https://github.com/thandal/CLIP_prefix_caption
from clip_prefix_caption.
Related Issues (20)
- it would be None
- Question about "clip_length" and "prefix_length" difference
- Some questions about fine-tune with custom dataset HOT 8
- AttributeError: module 'cog' has no attribute 'Predictor' HOT 3
- model overfitting issue HOT 5
- Parsing conceptual caption does not function properly as it removes some images and replaces them with zero tensor. HOT 1
- use different encoder HOT 3
- How to evaluate model with meteor, BLEU, or rouge HOT 3
- AttributeError: module 'cog' has no attribute 'Predictor' HOT 2
- Train costom data HOT 1
- Metrics of ClipCap's Original Performance HOT 2
- use multiple gpus to train
- How to evaluate the trained model? Is there a test.py ? HOT 6
- did anyone reproduce the transformer network with frozen GPT-2? HOT 7
- data json
- Where is the file 'model_wieghts.pt' exists?
- How to do eval, how to set the prompt
- How to inference after training on my own dataset HOT 1
- beamsearch lead to a worse result in inference script?
- Error in Load model weights HOT 3
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 clip_prefix_caption.