Thank you for your valuable work and for making the T31TFM-1618 dataset publicly available. It has been very helpful for our research on crop-type mapping using satellite time series.
In your paper, you mention that only crop types with more than 100k pixels are retained, resulting in 19 final crop classes used for training and evaluation.
However, neither the repository nor the paper provides the actual names of these 19 classes (e.g., “winter wheat”, “maize”, “sugar beet”, “permanent meadow”, etc.). Without this mapping, it is difficult to:
Interpret model performance per class,
Compare results with other datasets or studies,
Reproduce or extend your work meaningfully.
Could you please share:
The list of the 19 crop class names actually used in your experiments?
(If possible) The full list of original crop categories in the raw dataset before filtering? This would help us understand the aggregation or filtering process better.
We understand that the labels may originate from French agricultural records (e.g., RPG with CODE_CULTU), and any clarification—such as a simple table or text file mapping class indices (or codes) to human-readable names—would be immensely appreciated.
Thank you again for your contribution and for considering this request!
Thank you for your valuable work and for making the T31TFM-1618 dataset publicly available. It has been very helpful for our research on crop-type mapping using satellite time series.
In your paper, you mention that only crop types with more than 100k pixels are retained, resulting in 19 final crop classes used for training and evaluation.
However, neither the repository nor the paper provides the actual names of these 19 classes (e.g., “winter wheat”, “maize”, “sugar beet”, “permanent meadow”, etc.). Without this mapping, it is difficult to:
Interpret model performance per class,
Compare results with other datasets or studies,
Reproduce or extend your work meaningfully.
Could you please share:
The list of the 19 crop class names actually used in your experiments?
(If possible) The full list of original crop categories in the raw dataset before filtering? This would help us understand the aggregation or filtering process better.
We understand that the labels may originate from French agricultural records (e.g., RPG with CODE_CULTU), and any clarification—such as a simple table or text file mapping class indices (or codes) to human-readable names—would be immensely appreciated.
Thank you again for your contribution and for considering this request!