No, these are completely different, standalone code points, not variant forms of the same code point.
What's actually happening seems to be that the ch_tra model can recognize simplified too and output the corresponding traditional version if the character isn't in the traditional "alphabet"; it doesn't work so well in the other direction.
Example recognizing a partial screenshot of https://chinese.stackexchange.com/a/38707 (anyone can try this on Google Colab, no hardware required; remember to turn on GPU in Runtime -> Change runtime type):
import easyocr
import requests
zhs_reader = easyocr.Reader(['en', 'ch_sim'])
zht_reader = easyocr.Reader(['en', 'ch_tra'])
image = requests.get('https://i.imgur.com/HtrpZCZ.png').content
print('ch_sim:', ' '.join(text for _, text, _ in zhs_reader.readtext(image)))
print('ch_tra:', ' '.join(text for _, text, _ in zht_reader.readtext(image)))
Results:
ch_sim: One simplified character may mapping to multiple traditional ones: 皇后->皇后,後夭->后夭 豌鬟->头发,骏财->发财 As reversed, one traditional character may mapping to multiple simplified ones too: 乾燥->干燥, 乾隆->乾隆 嘹望->嘹望,嘹解->了解
ch_tra: One simplified character may mapping to multiple traditional ones: 皇后->皇后,後天->后天 頭髮->頭發,發財->發財 As reversed, one traditional character may mapping to multiple simplified ones too: 乾燥->干燥, 乾隆->乾隆 瞭望->瞭望, 瞭解->了解
Compare to the original text:
One simplified character may mapping to multiple traditional ones:
- 皇后 -> 皇后，後天 -> 后天
- 頭髮 -> 头发，發財 -> 发财
As reversed, one traditional character may mapping to multiple simplified ones too:
- 乾燥 -> 干燥，乾隆 -> 乾隆
- 瞭望 -> 瞭望，瞭解 -> 了解
Of course, automatic character-to-character conversion from simplified to traditional can be wrong due to ambiguities; excellent examples from above: 头发 => 頭發 (should be 頭髮), 了解 => 了解 (should be 瞭解).
This approach seems a bit weird to me. While I appreciate them separating the models of Traditional and Simplified Chinese, I think I might prefer them to be combined (perhaps even including Japanese Kanji), and instead provide a way for the user to specify which language or regional variant is expected so characters matching the expected variant are simply given a higher score.
Without delving into implementation details, I suspect the ch_tra model was simply trained on a dataset including simplified images with traditional labels.
