Classify an image

Submission guidelines

• Send the “closest” image to the source to the system: this means, for example, not sending screenshots or photos of screens.
• Image analysis models are not currently trained to recognize photos of documents: therefore, the result, in case of submission of images of that type, is not reliable.
• In the case of images taken from social media, the most accurate results are obtained by sending images without overlaid texts or modifications typical of social sharing.
• If available from other sources, avoid submitting social links due to the post-processing of various platforms.

Submit an image using the API

curl --request POST \
  --url https://backend.identifai.net/api/classification \
  --header 'Content-Type: multipart/form-data' \
  --header 'X-Api-Key: <api-key>' \
  --header 'Accept: application/json' \
  --form "image=@/path/to/sample.jpg" \
  --form with_morphing=false\
  --form with_tampering=false

Retrieve the results

curl --request GET \
  --url https://backend.identifai.net/api/classification/{identifier} \
  --header 'X-Api-Key: <api-key>'

Heatmap

curl --request POST \
  --url https://backend.identifai.net/api/classification \
  --header 'Content-Type: multipart/form-data' \
  --header 'X-Api-Key: <api-key>' \
  --header 'Accept: application/json' \
  --form "image=@/path/to/sample.jpg" \
  --form with_heatmap=true \
  --form with_morphing=false \
  --form with_tampering=false

{
  "results": {
      "model": "Ardē",
      "unknown": false,
      "elapsed_time": 5.606042385101318,
      "classification": [
        {
          "label": "human",
          "score": 0.8
        },
        {
          "label": "artificial",
          "score": 0.2
        }
      ]
    }
}

Smart Full Analysis

1. the image is split into a targeted selection of high-value crops;
2. each crop is analyzed and classified by the most advanced model available in the Revelio family;
3. the results of the single crops are combined into an overall verdict, taking into account the information collected from the different areas of the image.

curl --request POST \
  --url https://backend.identifai.net/api/classification \
  --header 'Content-Type: multipart/form-data' \
  --header 'X-Api-Key: <api-key>' \
  --header 'Accept: application/json' \
  --form "image=@/path/to/sample.jpg" \
  --form with_smart_analysis=true

XAI (eXplainable AI)

From classification to understanding

More transparency, more tools to interpret the results

Available on the web platform and via API

Enable XAI using the API

curl --request POST \
  --url https://backend.identifai.net/api/classification \
  --header 'Content-Type: multipart/form-data' \
  --header 'X-Api-Key: <api-key>' \
  --header 'Accept: application/json' \
  --form "image=@/path/to/sample.jpg" \
  --form with_xai=true \
  --form with_llm_xai=true \
  --form llm_xai_language=italian

XAI maps and explanations in the results

{
  "status": "done",
  "deferred_status": "done",
  "results": [
    {
      "model": "Revelio",
      "unknown": false,
      "elapsed_time": 1.234,
      "classification": [
        { "label": "human", "score": 0.03 },
        { "label": "artificial", "score": 0.97 }
      ],
      "explanations": [
        {
          "type": "attention_map",
          "method": "legrad",
          "description": "The model focuses on the skin texture around the eyes and on the hairline, where the generated details are less consistent."
        },
        {
          "type": "attention_graph",
          "method": "legrad",
          "description": null
        }
      ]
    }
  ]
}

• the AI-powered explanations were not requested (with_llm_xai not set): the maps are listed anyway, so that you can download them;
• the explanation is not available yet, because the deferred results are still in progress;
• the explanation was requested but could not be generated. If this happens because of a technical error, deferred_status is errored, while the explanations generated for the other maps are still returned: see Deferred results. The credits of the AI-powered explanations that cannot be generated are refunded.

• the explanations are returned only by the single classification result endpoint: they are not included in the lists of classifications nor in the payload sent to webhook_url. The notification sent to deferred_webhook_url contains the descriptions in its xai and heatmaps fields;
• the per-model results, and therefore the explanations, are returned only if your pricing plan exposes the details of the models;
• when the heatmap is requested, it is listed with type heatmap (and method null) on the result of the Ardē model; when the Smart Full Analysis is enabled, its heatmap is listed on the results of the models that produced it.

Download the Attention Map and the Attention Graph

curl --request GET \
  --url 'https://backend.identifai.net/api/classification/xai/{identifier}?model=Revelio&method=legrad&type=attention_map' \
  --header 'X-Api-Key: <api-key>' \
  --output attention_map.png

Guidelines on interpreting results

• Revelio as the main result for the verdict (platform’s default choice), with higher confidence if faces and people are present.
• Aide for a detailed analysis with the support of the heatmap feature.
• Morphing to be considered only in the presence of faces.
• Ellen, if the image comes from sources where the use of photo editing tools has been ascertained in the past.
• V01 and Brokenwand only in the case of very old images (and therefore potentially generated with systems no longer in use today).

• Pay close attention to the size of the submitted images: heatmap analysis is particularly demanding from a computational point of view and processing time increases significantly based on image size. The guideline recommends submitting images with a resolution lower than 2k (or 4 megapixels).

See also