Positive users focus on the VLM Judge Pipeline being orchestrated by Kimi K2.7 Code as an exciting test for the agent itself, while negative users highlight misjudging of coke bottles and chaotic labeling on fridge data with reflections.
Based on 2 visible X reactions from 2 accounts; directional sample.
Ask a question below.
Published answers will appear here.
I will let it run to check how the judges do though. the exciting thing is pipeline is orchestrated by Kimi K2.7 Code so it's a test for agent itself rather than pipeline
@mervenoyann it misjudging the coke bottles for sure though lol, labelling on fridge data always gets chaotic with reflections
just discovered a use case where my sideproject VLM as a judge pipeline could fail because what is this 😄 labelling is done with Qwen3.5-9B maybe I could just use a judge to generate boxes and take IoU or do NMS https://x.com/mervenoyann/status/2076097762764955814/photo/1 https://twitter.com/mervenoyann/status/2071712273069232503
I will let it run to check how the judges do though. the exciting thing is pipeline is orchestrated by Kimi K2.7 Code so it's a test for agent itself rather than pipeline
@mervenoyann it misjudging the coke bottles for sure though lol, labelling on fridge data always gets chaotic with reflections
just discovered a use case where my sideproject VLM as a judge pipeline could fail because what is this 😄 labelling is done with Qwen3.5-9B maybe I could just use a judge to generate boxes and take IoU or do NMS https://x.com/mervenoyann/status/2076097762764955814/photo/1 https://twitter.com/mervenoyann/status/2071712273069232503
Positive users focus on the VLM Judge Pipeline being orchestrated by Kimi K2.7 Code as an exciting test for the agent itself, while negative users highlight misjudging of coke bottles and chaotic labeling on fridge data with reflections.
Based on 2 visible X reactions from 2 accounts; directional sample.
Ask a question below.
Published answers will appear here.