Hi, I tried to run the evaluation code for refCOCO, refCOCO+ and refCOCOg with SPHINX-1k. The results were lower than the ones reported in the paper (see attached pic). What could be the reason here? I set the temperature to 0.1, top_p to 0.75 and no quantization was used.
Thanks for your great work !
Hi, I tried to run the evaluation code for refCOCO, refCOCO+ and refCOCOg with SPHINX-1k. The results were lower than the ones reported in the paper (see attached pic). What could be the reason here? I set the temperature to 0.1, top_p to 0.75 and no quantization was used.
Thanks for your great work !