search

marcelsheeny / radiate_sdk

SDK developed to access data from RADIATE dataset
MIT License
89 stars 32 forks source link

Provide Training/Testing Splits #30

Open soulertsai opened 2 years ago

soulertsai commented 2 years ago

I didn't find how the evaluation is done and Training/Testing Splits, which is hard to do benchmarking. Is it possible to provide Training/Testing Splits information in the paper?

I have the following folders after downloading the dataset, and I found the total number of imgs is different from the sum of training and testing num in the paper.

city_1_0
city_3_1
city_3_2
city_3_3
city_3_7
city_4_0
city_5_0
city_6_0
city_7_0
fog_6_0
fog_8_2
junction_1_0
junction_1_10
junction_1_11
junction_1_12
junction_1_14
junction_1_4
junction_1_6
junction_1_7
junction_1_8
junction_1_9
junction_2_0
junction_2_1
junction_2_2
junction_2_5
junction_2_6
junction_3_1
junction_3_2
junction_3_3
motorway_1_0
motorway_2_0
motorway_2_1
motorway_2_2
night_1_2
night_1_5
rain_2_0
rain_3_0
rain_4_0
rural_1_1
snow_1_0

Related To: https://github.com/marcelsheeny/radiate_sdk/issues/23

Madhusakth commented 2 years ago

Hi, If I understand your question, you are asking about the test and train split. If that is correct, in each folder, there is a meta.json file that specifies if this set is test/train_good_weather/train_good_bad_weather. I hope this helps!