Comments (16)
Yes I have managed to train and evaluate the model on my own custom dataset, I don't have test set in my dataset, I have only train and val set. so the test.yaml file points to the val set.
img: val.img.tsv
hw: val.hw.tsv
label: val.label.tsv
feature: val.feature.tsv
caption: val_caption.json
this is my test.yaml and val.yaml fil content.
But first you need to create the .tsv and .json files. If you need help in this regard I can point you the issues which helped me to prepare those files
from oscar.
We have saved all running scripts to: https://github.com/microsoft/Oscar/blob/master/MODEL_ZOO.md
Is this what you are looking for?
from oscar.
the train_yaml is from our dataset, you can download it: https://github.com/microsoft/Oscar/blob/master/DOWNLOAD.md
from oscar.
To get the image captioning dataset, run this:
wget https://biglmdiag.blob.core.windows.net/oscar/datasets/coco_caption.zip
Then train.yaml is a yaml file specifies the data like features and labels used in training. It comes with the dataset.
from oscar.
MIN_BOXES=0, MAX_BOXES=100
conf_thresh = 0.2 or 0.4, I forget the exact value, great probability it is 0.2
from oscar.
The run_captioning.py file has the train_yaml config but I have not found this.
from oscar.
The file for generating the downloaded files?
from oscar.
Recently, I tried to generate the COCO caption feature files according to your format. The model you used is the bottom-up-attention,right? https://github.com/peteanderson80/bottom-up-attention
If so, there exists some hyperparameters that I can not understand. I found most of the images in you feature tsv contain 20~30 bounding boxes. So if convenient,could you share the MIN_BOXES , MAX_BOXES as well as the conf_thresh hyperparameter?
from oscar.
The bottom up model is trained in Visual Genome dataset,right?
from oscar.
Yes
from oscar.
I want to test the model with my own images and now I generated the features and labels. however when I run the script oscar/run_captioning.py It needs test.yaml. Do I have to download the whole training data to test the model on my own images?
This is what I run on my terminal:
python oscar/run_captioning.py
--do_test --do_eval --test_yaml test.yaml
--per_gpu_eval_batch_size 64 --num_beams 5
--max_gen_length 20 --eval_model_dir oscar/checkpoint-29-66420
from oscar.
I want to test the model with my own images and now I generated the features and labels. however when I run the script oscar/run_captioning.py It needs test.yaml. Do I have to download the whole training data to test the model on my own images?
This is what I run on my terminal:
python oscar/run_captioning.py
--do_test --do_eval --test_yaml test.yaml
--per_gpu_eval_batch_size 64 --num_beams 5
--max_gen_length 20 --eval_model_dir oscar/checkpoint-29-66420
@DesaleF Were you able to figure out the test.yaml file to test the model on your own images? I'm running into the same issue with not knowing how create test.yaml
from oscar.
Can some one tell me how to get only a part of the dataset for training ? 22GB to huge. Also, how to do inference on a downloaded pretrained model ? how to get the test.yaml file ?
from oscar.
Hey it would be really nice if you could share those links or an example of what you created. I need to do the same :/
from oscar.
Can some one tell me how to get only a part of the dataset for training ? 22GB to huge. Also, how to do inference on a downloaded pretrained model ? how to get the test.yaml file ?
Also just to confirm, do we need the whole download to run inference? I am not sure if the tokenizer is already in the repo or if its in that massive download.
from oscar.
Yes I have managed to train and evaluate the model on my own custom dataset, I don't have test set in my dataset, I have only train and val set. so the test.yaml file points to the val set.
img: val.img.tsv hw: val.hw.tsv label: val.label.tsv feature: val.feature.tsv caption: val_caption.json
this is my test.yaml and val.yaml fil content. But first you need to create the .tsv and .json files. If you need help in this regard I can point you the issues which helped me to prepare those files
How to get label.tsv? Thanks!
from oscar.
Related Issues (20)
- Why do I need the test_caption.json file to generate caption
- Some doubt about contrastive loss and the output of BertImgForPreTraining HOT 1
- what is checkpoint-29-66420 HOT 2
- How to use Oscar / VinVL for image-text retreival inference?
- Object tag dictionary for OSCAR (vanilla)
- How are the files in "train_logs/" and "test_logs/" encoded? HOT 1
- missing "imageid2idx.json" when run script run_retrieval.py
- VinVL features for datasets not available HOT 2
- Can you share the full NoCaps results on the test data? HOT 1
- About checkpoint
- The specified resource does not exist. HOT 5
- Vocabulary of the test split
- How do you implement the multi-layer transformers? HOT 1
- ModuleNotFoundError: No module named 'oscar' HOT 1
- Public access for pretrained models are blocked HOT 5
- Azcopy failed even with SAS token in the VinVL_DOWNLOAD.md file HOT 8
- Unable to run Oscar for Image Captioning
- Oscar fails with torch 2.1
- Azcopy failes with given SAS token HOT 1
- "https://biglmdiag.blob.core.windows.net/vinvl/datasets/coco_caption" file does not download
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 oscar.