Official repository of OFA (ICML 2022). Paper: OFA: Unifying Architectures, Tasks, and Modalities Through a Simple Sequence-to-Sequence Learning Framework
Apache License 2.0
2.39k
stars
248
forks
source link
How to obtain Accuracy for Visual Grounding after running evaluation code #385
So this should be a quick one. If anybody has pointers or suggestions please let me know.
I have finished fine-tuning OFA for Visual Grounding and I ran the evaluation scripts. After running the evaluation scripts, JSON files for the eval data were generated and stored in a result folder like the one below:
And when I open one of them, I notice that they are a collection of image IDs and bounding boxes.
Now I need to obtain the accuracy as shown on the OFA GitHub homepage. I thought there should be a script to obtain the accuracies but I cannot find it. If anyone knows where I can find the scripts please let me know. On the other hand, how have people computed accuracy? Should I just come up with a script by myself?
So this should be a quick one. If anybody has pointers or suggestions please let me know.
I have finished fine-tuning OFA for Visual Grounding and I ran the evaluation scripts. After running the evaluation scripts, JSON files for the eval data were generated and stored in a result folder like the one below:
And when I open one of them, I notice that they are a collection of image IDs and bounding boxes.
{ "uniq_id": "751_0", "box": [ 113.42108154296875, 44.214996337890625, 335.1368713378906, 265.28997802734375 ] }, { "uniq_id": "751_1", "box": [ 113.42108154296875, 44.214996337890625, 335.1368713378906, 265.28997802734375 ] }
Now I need to obtain the accuracy as shown on the OFA GitHub homepage. I thought there should be a script to obtain the accuracies but I cannot find it. If anyone knows where I can find the scripts please let me know. On the other hand, how have people computed accuracy? Should I just come up with a script by myself?