Meme detail

Text detected in the image

TowakiTakikawa/@yongyuanxi·154 Adding horizontal lines to images improves VLM (vision language model) performance of tasks like counting, visual search, spatial understating, scene understanding, and more How many Circles are there in this image? Scan the image sequentially based on horizontal lines 10 exist in the image. Getting Response Raw Image Applying row-wise structure Special prompt for task Table I: Harmonic mean comparison for the Visual Search task, evaluating our method against base models across GPT-4o, Claude3.5-sonnet, LLaMa4, and Qwen2.5-VL on 2D and 3D datasets with varying numbers of objects. GPT-40 Claude-sonnet LLaMa4 Qwen25-VL Dataset Objects Simple Ours Simple Ours Simple Ours Simple Ours Visual Search Counting Spatial Relationship 2D 20 0.71 0.91 0.55 0.82 0.00 0.06 0.31 0.43 20 0.50 0.67 20 0.12 20 0.41 0.50 0.80 0.54 0.70 0.00 0.00 0.28 0.37 0.73 0.25 0.39 0.43 40 0.25 0.64 0.18 0.59 0.00 0.00 0.14 0.23 0.33 50 0.46 0.55 0.10 0.55 0.00 0.00 0.47 0.56 0.92 0.95 0.15 063 0.64 Avg 0.48 0.73 0.34 0.66 0.00 0.02 0.30 0.40 30 0.94 30 0.05 3D 20 1.00 1.00 1.00 1.00 0.82 0.80 0.33 0.44 0.31 0.69 30 0.96 0.96 0.88 0.91 0.43 0.60 0.15 0.18 0.6 Harmonic Mean 0.8 1.0 0.1 AcC 0.2 0.3 0.4 0.4 03 ACCUCy 0.6 0.7 40 0.89 0.94 0.72 0.80 0.59 0.58 0.00 0.15 50 0.81 0.84 0.59 0.75 0.39 0.38 0.00 0.04 Avg 0.91 0.93 0.80 0.86 0.56 0.59 0.12 0.20

Popularity