Update README.md

README.md

@@ -13,10 +13,15 @@ The benchmark consists of four splits:
 The **Existence** split evaluates the object recognition ability while the remaining splits are designed to evaluate the capability of relation comprehension, focusing on probing each of the elements in the subject-predicate-object triplets of the scene graph separately.
 Some data examples are shown below.
 
-![crpe.jpg](https://cdn-uploads.huggingface.co/production/uploads/619507e7b74b6c591f794340/_NKaowl2OUBAjck1XCAPm.jpeg)
+<img width="800" alt="image" src="https://cdn-uploads.huggingface.co/production/uploads/619507e7b74b6c591f794340/_NKaowl2OUBAjck1XCAPm.jpeg">
+
+Additionally, to evaluate the dependency on language priors, we also include abnormal data in our evaluation.
+These images in these abnormal data depict relation triplets that are very rare in the real world.
+
+<img width="800" alt="image" src="https://cdn-uploads.huggingface.co/production/uploads/619507e7b74b6c591f794340/qKWw7Qb93OXClxI_VrCRk.jpeg">
 
 For a robust evaluation, we adopt CircularEval as our evaluation strategy.
 Under this setting, a question is considered as correctly answered only when the model consistently predicts the correct answer in each of the N iterations, with N corresponding to the number of choices.
 In each iteration, a circular shift is applied to both the choices and the answer to form a new query for the model.
 
-See our [paper](https://github.com/OpenGVLab/all-seeing/all-seeing-v2) to learn more details!
+See our [project](https://github.com/OpenGVLab/all-seeing/all-seeing-v2) to learn more details!