Commit 623f1b68 authored by Jiang, Genze's avatar Jiang, Genze
Browse files

Update metadata.json

parent 951e74b4
......@@ -30,7 +30,7 @@
"imgTexts": "when Japan thinks you are gonna invade their mainland",
"clsResult": "Your uploaded image and text combination looks like a <strong>HATEFUL</strong> meme, with 94.79% confidence.",
"shap": {
"modelType": "MMBT",
"modelType": "LateFusion",
"imgExp": "examples/40375/40375_shap_img.png",
"txtExp": "examples/40375/40375_shap_txt.png",
"imgMsg": "<p><strong>tl;dr:</strong><br> <span style=\"color: red\">Red</span> (<span style=\"color: blue\">Blue</span>) regions move the model output towards Hateful (Non-hateful).</p><p><strong>Details</strong>:<br>The input image is segmented into 52 regions, and text string is split into 9 features. The shapley values of those 61 features represent their additive contributions towards the model output for the current inclination selected, 0.948, on top of the base value. The base value, 0.0374, is the expected model output without those features. The sum of all shapley values and the base value should equate the selected model output, i.e.</p><p><em>model_output = base_value + total_image_shapley_values + total_text_shapley_values</em>.</p><p>The sum of shapley values for the image features is 0.5587.</p><span style=\"font-size: 0.75rem\">*note that the results may change slightly if the number of evaluations is small due to random sampling in the algorithm</span>",
Supports Markdown
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment