# FastVLM-0.5B 能读懂哪些图片？24 张实测 在浏览器测试收据、图表、界面、英文小字、中文海报和形状计数。公开全部原图、固定问题、预期答案与未编辑的模型输出。 Canonical URL: https://aiidelist.com/zh/blog/fastvlm-image-tests-zh Language: zh Published: 2026-09-24 Updated: 2026-09-24 FASTVLM-0.5B · SITE IMAGE TESTS 在浏览器测试收据、图表、界面、英文小字、中文海报和形状计数。公开全部原图、固定问题、预期答案与未编辑的模型输出。 测量日期：2026-09-12 · 全部案例公开 **答案精确匹配** 8/24 **包含预期答案** 21/24 **图片 → 回答耗时中位数** 732.5 ms 这是小型合成集上的回答格式与目标出现检查，不是通用 OCR 准确率。回答包含目标时，仍可能附带错误信息。 ## 测试了什么 主机为 Apple M5 Pro、48 GiB 统一内存，使用无头 Chrome 152.0.7977.83 与 WebGPU。模型为 onnx-community/FastVLM-0.5B-ONNX，Transformers.js 3.7.2；视觉编码器与解码器 q4、词嵌入 fp16。先用独立收据预热一次，再对每张图运行一次。输入 800×600，页面在模型预处理前将最长边限制为 1024 像素。贪心生成，最多新增 192 个 token，重复惩罚 1.2。 耗时从图片处理前开始，到生成结束，不含模型准备、选择图片与导出 JSON。原始数据中的首段文字计时截止于第一个非空文本片段，不是严格的首 token 事件。每张图的一次计时不足以证明跨设备的稳定性能。 24 张图片均为原创合成图，由六类模板、每类四个变体构成。不包含照片、扫描件、手写内容或用户真实文档。大部分问题只问一个字段或类别，并非全文转录。字体、布局与标签在推理前已固定；这个集合不足以用于通用模型排名。 ## 按任务类别查看结果 | 类别 | 答案精确匹配 | 包含预期答案 | 图片 → 回答耗时中位数 | | --- | --- | --- | --- | | 收据 | 4/4 | 4/4 | 719.0 ms | | 图表 | 0/4 | 4/4 | 697.5 ms | | 界面 | 0/4 | 4/4 | 1884.5 ms | | 英文小字 | 0/4 | 4/4 | 746.0 ms | | 中文 | 0/4 | 1/4 | 971.0 ms | | 形状计数 | 4/4 | 4/4 | 699.5 ms | 精确匹配会去掉首尾空白、一个前置美元符号和一个末尾标点，再忽略英文大小写比较。“包含预期答案”对英文和数字使用完整 token，对中文使用子串检查。两项均不判断附加解释是否真实。 ## 这些案例说明什么 - 四个收据总额和四个蓝色圆形数量都严格匹配了要求的回答。 - 图表名称、界面状态和英文产品编号出现在回答中，但模型经常在要求简答时添加解释，后续程序不能默认它遵守输出格式。 - 四道中文地点题中，三道没有出现图上地点。剩下一道虽包含预期地点，也附加了没有依据的说法。此配置用于中文提取时需要人工核对。 可以用 Playground 测试自己的任务并检查输出。目前证据支持有人检查的图片问答体验，尚不足以把它作为完全可靠、无需校对的 OCR 工具。 ## 自行复现 [下载完整测试包 ↓](/fastvlm/evaluations/2026-09-12/evaluation-kit.zip) · [试试自己的图片 →](/zh/fastvlm/demo)Chrome + a running FastVLM site · model download required ```text npm install --save-dev @playwright/test@1.55.1 # First start the FastVLM site locally; the script uses its English /demo. PLAYWRIGHT_BASE_URL=http://127.0.0.1:4178 node scripts/evaluate-browser.mjs ``` [原始回答与耗时](/fastvlm/evaluations/2026-09-12/fastvlm-webgpu.json) · [Manifest](/fastvlm/evaluations/2026-09-12/manifest.json) · [Provenance](/fastvlm/evaluations/2026-09-12/provenance.json) · [查看相同图片上的 MobileCLIP2 S0 / S2 对比](/zh/blog/mobileclip2-s0-vs-s2-zh) ## 逐例查看全部结果 类别 仅显示缺失预期答案 / 分类错误 24 个案例 “包含预期答案”只检查目标是否出现，回答其余部分仍可能编造信息。“精确匹配”同时检验是否遵循了简短回答的格式要求。 [![receipt-1](/fastvlm/evaluations/2026-09-12/images/receipt-1.png)](/fastvlm/evaluations/2026-09-12/images/receipt-1.png) ### receipt-1 **问题:** What is the TOTAL amount? Return the amount only. **预期答案:** 15.00 **实际回答** $15.00 **精确匹配** 是 **包含预期答案** 是 **图片 → 回答** 720 ms [![receipt-2](/fastvlm/evaluations/2026-09-12/images/receipt-2.png)](/fastvlm/evaluations/2026-09-12/images/receipt-2.png) ### receipt-2 **问题:** What is the TOTAL amount? Return the amount only. **预期答案:** 27.50 **实际回答** $27.50 **精确匹配** 是 **包含预期答案** 是 **图片 → 回答** 718 ms [![receipt-3](/fastvlm/evaluations/2026-09-12/images/receipt-3.png)](/fastvlm/evaluations/2026-09-12/images/receipt-3.png) ### receipt-3 **问题:** What is the TOTAL amount? Return the amount only. **预期答案:** 7.25 **实际回答** $7.25 **精确匹配** 是 **包含预期答案** 是 **图片 → 回答** 691 ms [![receipt-4](/fastvlm/evaluations/2026-09-12/images/receipt-4.png)](/fastvlm/evaluations/2026-09-12/images/receipt-4.png) ### receipt-4 **问题:** What is the TOTAL amount? Return the amount only. **预期答案:** 100.90 **实际回答** $100.90 **精确匹配** 是 **包含预期答案** 是 **图片 → 回答** 723 ms [![chart-1](/fastvlm/evaluations/2026-09-12/images/chart-1.png)](/fastvlm/evaluations/2026-09-12/images/chart-1.png) ### chart-1 **问题:** Which category has the highest number of orders? Answer with its name only. **预期答案:** Beta **实际回答** Answer: Beta **精确匹配** 否 **包含预期答案** 是 **图片 → 回答** 694 ms [![chart-2](/fastvlm/evaluations/2026-09-12/images/chart-2.png)](/fastvlm/evaluations/2026-09-12/images/chart-2.png) ### chart-2 **问题:** Which category has the highest number of orders? Answer with its name only. **预期答案:** Alpha **实际回答** Answer: Alpha **精确匹配** 否 **包含预期答案** 是 **图片 → 回答** 701 ms [![chart-3](/fastvlm/evaluations/2026-09-12/images/chart-3.png)](/fastvlm/evaluations/2026-09-12/images/chart-3.png) ### chart-3 **问题:** Which category has the highest number of orders? Answer with its name only. **预期答案:** Gamma **实际回答** Answer: Gamma **精确匹配** 否 **包含预期答案** 是 **图片 → 回答** 685 ms [![chart-4](/fastvlm/evaluations/2026-09-12/images/chart-4.png)](/fastvlm/evaluations/2026-09-12/images/chart-4.png) ### chart-4 **问题:** Which category has the highest number of orders? Answer with its name only. **预期答案:** Beta **实际回答** Answer: Beta **精确匹配** 否 **包含预期答案** 是 **图片 → 回答** 717 ms [![interface-1](/fastvlm/evaluations/2026-09-12/images/interface-1.png)](/fastvlm/evaluations/2026-09-12/images/interface-1.png) ### interface-1 **问题:** What is the delivery status of order #1042? Return the status only. **预期答案:** Pending **实际回答** To determine the delivery status of Order #1042, we need to carefully examine the provided image. The image shows an interface with various elements such as buttons for "View" and "Order." Step 1: Identify the relevant information. - Locate the section labeled "Delivery status." - Observe that there is a button next to this label. Step 2: Read the text on the button. - The button clearly states "Pending." Therefore, based on the information in the image, the delivery status of Order #1042 is "Pending." This indicates that the order has not yet been delivered or processed by the system. **精确匹配** 否 **包含预期答案** 是 **图片 → 回答** 1563 ms [![interface-2](/fastvlm/evaluations/2026-09-12/images/interface-2.png)](/fastvlm/evaluations/2026-09-12/images/interface-2.png) ### interface-2 **问题:** What is the delivery status of order #1042? Return the status only. **预期答案:** Approved **实际回答** To determine the delivery status of order #1042, we need to carefully examine the provided image. The image shows an interface with various elements such as buttons, text fields, and links. Step 1: Identify the relevant information. - In this case, we are looking for the "Delivery status" of order #1042. Step 2: Locate the button labeled "Approved." - Upon examining the image, we see that there is a blue button labeled "Approved." Step 3: Confirm the delivery status based on the label. - Since the button labeled "Approved" indicates that the delivery status has been approved, we can conclude that the delivery status of order #1042 is indeed approved. Therefore, the delivery status of order #1042 is Approved. **精确匹配** 否 **包含预期答案** 是 **图片 → 回答** 1802 ms [![interface-3](/fastvlm/evaluations/2026-09-12/images/interface-3.png)](/fastvlm/evaluations/2026-09-12/images/interface-3.png) ### interface-3 **问题:** What is the delivery status of order #1042? Return the status only. **预期答案:** Shipped **实际回答** To determine the delivery status of order #1042, we need to carefully examine the provided image. The image shows an interface with various elements such as buttons for "View" and "Order". Step 1: Identify the relevant information. - In this case, we focus on the button labeled "Shipped". Step 2: Understand the context. - The text above the "Shipped" button reads "Order #1042", indicating that this is the specific order being referred to. Step 3: Confirm the delivery status based on the given instruction. - According to the instructions in the simple answer, the delivery status should be specified by clicking or hovering over the "Shipped" button. Therefore, the delivery status of order #1042 can be determined by following these steps: 1. Locate the "Shipped" button in the image. 2. Read the accompanying text which states "Order #10 **精确匹配** 否 **包含预期答案** 是 **图片 → 回答** 1967 ms [![interface-4](/fastvlm/evaluations/2026-09-12/images/interface-4.png)](/fastvlm/evaluations/2026-09-12/images/interface-4.png) ### interface-4 **问题:** What is the delivery status of order #1042? Return the status only. **预期答案:** Cancelled **实际回答** To determine the delivery status of Order #1042, we need to carefully examine the provided image. The image shows an interface with various elements such as buttons for "View" and "Order," along with text labels. Step-by-Step Solution: 1. **Identify the Relevant Information:** - Locate the section labeled "Delivery status." - Within this section, there is a button that reads "Cancelled." 2. **Analyze the Button's Text:** - The button has the label "Cancelled" which indicates the current status of the order. 3. **Conclusion:** - Based on the information provided in the image, the delivery status of Order #1042 is "Canceled." Therefore, the detailed answer is: The delivery status of Order #1042 is canceled. This conclusion can be drawn by examining the relevant section of the interface where the "Order Status" option is highlighted, **精确匹配** 否 **包含预期答案** 是 **图片 → 回答** 2059 ms [![small-text-1](/fastvlm/evaluations/2026-09-12/images/small-text-1.png)](/fastvlm/evaluations/2026-09-12/images/small-text-1.png) ### small-text-1 **问题:** What is the product code? Return the code only. **预期答案:** AX-4179 **实际回答** The product code is AX-4179. **精确匹配** 否 **包含预期答案** 是 **图片 → 回答** 783 ms [![small-text-2](/fastvlm/evaluations/2026-09-12/images/small-text-2.png)](/fastvlm/evaluations/2026-09-12/images/small-text-2.png) ### small-text-2 **问题:** What is the product code? Return the code only. **预期答案:** BK-5826 **实际回答** The product code is BK-5826. **精确匹配** 否 **包含预期答案** 是 **图片 → 回答** 744 ms [![small-text-3](/fastvlm/evaluations/2026-09-12/images/small-text-3.png)](/fastvlm/evaluations/2026-09-12/images/small-text-3.png) ### small-text-3 **问题:** What is the product code? Return the code only. **预期答案:** CM-9034 **实际回答** The product code is CM-9034. **精确匹配** 否 **包含预期答案** 是 **图片 → 回答** 748 ms [![small-text-4](/fastvlm/evaluations/2026-09-12/images/small-text-4.png)](/fastvlm/evaluations/2026-09-12/images/small-text-4.png) ### small-text-4 **问题:** What is the product code? Return the code only. **预期答案:** DT-2681 **实际回答** The product code is DT-2681. **精确匹配** 否 **包含预期答案** 是 **图片 → 回答** 742 ms [![chinese-1](/fastvlm/evaluations/2026-09-12/images/chinese-1.png)](/fastvlm/evaluations/2026-09-12/images/chinese-1.png) ### chinese-1 **问题:** 活动地点在哪里？只回答地点名称。 **预期答案:** 城市图书馆 **实际回答** 根据提供的文本，该活动的地点是“中国北京”。“中国北京”的位置位于中国首都北京，具体来说是在东城区，与周边地区相对应。这可以从所提供的地理信息中推断出来，因为提到的是“一座城市和一个国家”。 **精确匹配** 否 **包含预期答案** 否 **图片 → 回答** 1056 ms [![chinese-2](/fastvlm/evaluations/2026-09-12/images/chinese-2.png)](/fastvlm/evaluations/2026-09-12/images/chinese-2.png) ### chinese-2 **问题:** 活动地点在哪里？只回答地点名称。 **预期答案:** 社区活动中心 **实际回答** 根据图片中提供的信息，该活动的地点是“中国云南”(China Yunnan). 这些文本表明了地理位置和文化背景，但没有提供具体位置或详细说明。 **精确匹配** 否 **包含预期答案** 否 **图片 → 回答** 980 ms [![chinese-3](/fastvlm/evaluations/2026-09-12/images/chinese-3.png)](/fastvlm/evaluations/2026-09-12/images/chinese-3.png) ### chinese-3 **问题:** 活动地点在哪里？只回答地点名称。 **预期答案:** 人民公园 **实际回答** 根据图片中提供的信息，该活动的地点是“人民公园”。这可以通过文本中的第二行所示，“市民公园”被翻译为“人民公园”，表明它是一个公共区域或公园。 **精确匹配** 否 **包含预期答案** 是 **图片 → 回答** 962 ms [![chinese-4](/fastvlm/evaluations/2026-09-12/images/chinese-4.png)](/fastvlm/evaluations/2026-09-12/images/chinese-4.png) ### chinese-4 **问题:** 活动地点在哪里？只回答地点名称。 **预期答案:** 科学博物馆 **实际回答** 根据提供的文本，该活动的地点是“中国云南三元街”。这表明该活动发生在三元街，这是中国云南的一条主要街道和文化街区之一。 **精确匹配** 否 **包含预期答案** 否 **图片 → 回答** 948 ms [![shapes-1](/fastvlm/evaluations/2026-09-12/images/shapes-1.png)](/fastvlm/evaluations/2026-09-12/images/shapes-1.png) ### shapes-1 **问题:** How many blue circles are in the image? Return one digit only. **预期答案:** 1 **实际回答** 1 **精确匹配** 是 **包含预期答案** 是 **图片 → 回答** 708 ms [![shapes-2](/fastvlm/evaluations/2026-09-12/images/shapes-2.png)](/fastvlm/evaluations/2026-09-12/images/shapes-2.png) ### shapes-2 **问题:** How many blue circles are in the image? Return one digit only. **预期答案:** 2 **实际回答** 2 **精确匹配** 是 **包含预期答案** 是 **图片 → 回答** 678 ms [![shapes-3](/fastvlm/evaluations/2026-09-12/images/shapes-3.png)](/fastvlm/evaluations/2026-09-12/images/shapes-3.png) ### shapes-3 **问题:** How many blue circles are in the image? Return one digit only. **预期答案:** 3 **实际回答** 3 **精确匹配** 是 **包含预期答案** 是 **图片 → 回答** 696 ms [![shapes-4](/fastvlm/evaluations/2026-09-12/images/shapes-4.png)](/fastvlm/evaluations/2026-09-12/images/shapes-4.png) ### shapes-4 **问题:** How many blue circles are in the image? Return one digit only. **预期答案:** 4 **实际回答** 4 **精确匹配** 是 **包含预期答案** 是 **图片 → 回答** 703 ms