By skyforbes Hello – I'm trying Gemini for OCR (it's simpler to use than the API) but it makes clear changes the text.
This is a fragment of an image that I gave to Flash 2.5:
The prompt was "OCR this, without making any changes to the text"
It took some liberties with the output, removing a pair of # marks near the end.
multi_entry_queries = {}
for idx, entry_str in enumerate(stream, start=1):
try:
entry = json.loads(entry_str)
except Exception as e:
print(f"Cannot parse #{idx}: {entry_str.strip()}", file=sys.stderr)
continue
if 'c' in entry and entry['c'] == "COMMAND":
continue
if '$' in entry.get('ns', ''):
continue
Another example:
# "create" goes to position 0, etc
crud_counts[ts_min_s][crud_index(cmd_type[0])] += 1
These look like attempts at correction (it recognized and highlighted the Python code). How do I get it to be faithful? I've tried including clauses like "don't make any changes"; "don't interpret the code"; "recover the original text". It won't "listen".
Any ideas?