metadata

license: apache-2.0
datasets:
  - arampacha/rsicd
metrics:
  - bleu
library_name: transformers
pipeline_tag: image-to-text

BLIP for RSICD image captioning:

blip-image-captioning-base model has been finetuned on the rsicd dataset. Training parameters used are as follows:
- learning_rate = 5e-7
- optimizer = AdamW
- scheduler = ReduceLROnPlateau
- epochs = 5
More details (demo, testing, evaluation, metrics) available at github repo