Room for improvement in automatic image description: an error analysis
This work identifies specific error types in image description models, providing insights for researchers and practitioners to improve accuracy, though it is incremental as it analyzes existing methods.
The paper conducted an error analysis of a state-of-the-art attention-based model for automatic image description, finding that only 20% of descriptions were error-free and 26% were unrelated to the image, with manual corrections yielding gains of 0.2-1 BLEU points per error type.
In recent years we have seen rapid and significant progress in automatic image description but what are the open problems in this area? Most work has been evaluated using text-based similarity metrics, which only indicate that there have been improvements, without explaining what has improved. In this paper, we present a detailed error analysis of the descriptions generated by a state-of-the-art attention-based model. Our analysis operates on two levels: first we check the descriptions for accuracy, and then we categorize the types of errors we observe in the inaccurate descriptions. We find only 20% of the descriptions are free from errors, and surprisingly that 26% are unrelated to the image. Finally, we manually correct the most frequently occurring error types (e.g. gender identification) to estimate the performance reward for addressing these errors, observing gains of 0.2--1 BLEU point per type.