Product Detection

Extract a specific retail product with the reference image in the video with timestamps.

You can query a model directly and test the results returned when using different parameter values with the Cloud console, or by calling the Vertex AI API directly.

System instructions

You are a video data analyst with expertise in commercial products. Please do not hallucinate. You can just output nothing if there are no positive findings. Do not output findings for products not matching the reference image.

Prompt Freeform

Task description: You are given a retail product's image and a video to label. Your task is to examine every video frame and find out all the occurrences of this reference product in each frame. Explain your finding for each occurrence.

Output specification: You should provide the output in a strictly valid JSON format same as the following example.

[
  {
    "timestamp": "Timestamp of the event in mm:ss format.",
    "reason": "How does the product appear in the video frame."
  },
  {
    "timestamp": "Timestamp of the event in mm:ss format.",
    "reason": "How does the product appear in the video frame."
  },
]

Be aware that the start and end time must be in a strict numeric format: mm:ss. Do not output anything after the JSON content. If there are no matching logos, you should not output anything. Please do not hallucinate. Your answer (as a JSON LIST):

Response

[
  {
    "timestamp": "00:05",
    "reason": "The product is shown in a partial, out-of-focus view."
  },
  {
    "timestamp": "00:08",
    "reason": "The product is shown in a partial, out-of-focus view."
  },
  {
    "timestamp": "00:13",
    "reason": "The product is shown in a partial, out-of-focus view."
  },
  {
    "timestamp": "00:16",
    "reason": "The product is shown in a partial, out-of-focus view."
  },
  {
    "timestamp": "00:18",
    "reason": "The product is shown in a partial, out-of-focus view."
  },
  {
    "timestamp": "00:29",
    "reason": "The product is featured in a close-up shot, showcasing its design and functionality."
  },
  {
    "timestamp": "00:33",
    "reason": "The product is featured in a close-up shot, showcasing its design and functionality."
  },
  {
    "timestamp": "00:36",
    "reason": "The product is featured in a close-up shot, showcasing its design and functionality."
  },
  {
    "timestamp": "00:43",
    "reason": "The product is featured in a close-up shot, showcasing its design and functionality."
  }
]
Model: gemini-2.5-pro
Temperature: 0.2
Max output tokens: 8192
TopK: 40.0
TopP: 0.95