search

BubblyYi / MMPedestron

[ECCV2024] Official implementation of the paper "When Pedestrian Detection Meets Multi-Modal Learning: Generalist Model and Benchmark Dataset"
Other
47 stars 3 forks source link

Quality metrics on CrowdHuman are lower than in the paper #10

Open PredelinaAsya opened 2 weeks ago

PredelinaAsya commented 2 weeks ago

Good afternoon!

Faced with problem: when running the model on CrowdHuman validation, the metrics are lower than they should be.

I tried to run it in the following ways:

  1. the general model with the config (I left only the part with CrowdHuman in the dataset) and weights gave results:

    mmpedestron_metrics_pretrained_on_crowdhuman

  2. CrowdHuman model`s with the config and weigths results:

    mmpedestron_crowdhuman_model_metrics

The paper presents the following metrics:

paper_mmpedestron_metrics_on_crowdhuman

@BubblyYi @jin-s13 Please, explain how you ran the model on CrowdHuman to get the metrics described in the paper? Maybe, I'm doing something wrong.