Skip to content

The result using this code base is much lower than that showed in the paper #6

Description

@jerry-lcj

Hi, I am trying to reproduce the result of this code base, I use cholect50 as data and follow the data split instructions. The train.sh is used as follow:
#! /bin/bash

CUDA_VISIBLE_DEVICES=0, python3 main.py
--exp_name rit_train
--log_name rit_train.log
--max_epochs 40
--early_stopping_patience 5
--fold 1
--split 'cholect45-crossval'
--m 6
--bs 16
--nw 6
--topK 5
--ln 0
--cg 0
--od1 1e-6
--od2 1e-6
--od3 1e-6
--mom 0.95
--ms1 20
--ms2 39
--ms3 60
--g1 0.94
--g2 0.95
--g3 0.99
--layers 8
--data_dir "path-to-CholecT50" \

Plus I also change the learning rate to 1e-3 to avoid the issue of under fitting. But i only got the IVT mAP as 0.1. Is there any training details i missed?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions