TextCaps #dataset #機械学習 #todo [text captioning]用のデータセット [text captioning]: OCR的側面を持ったImage Captioning 👇COCO+既存モデルだとこのタスクは解けない👇 評価はどうするの問題 We report BLEU, METEOR, ROUGE-L, SPICE and CIDEr metrics for each submissions using coco-captions package. The leaderboard scores are sorted based on CIDEr and final ranking will be provided after the challenge based on a mixture of human evaluations and CIDEr. 引用: https://textvqa.org/textcaps/challenge/ どうやら, n-gram評価系列ではないSPICE等も評価に使うらしい [SPICE: ...