MiniCPM-V-4.7-1B-RKNN3-RK1828 / host /make_test_images.py
jj90's picture
Add model card, deploy guide, host + conversion scripts
fd11f30 verified
Raw History Blame Contribute Delete
1.94 kB
"""生成两张 448x448 测试图(纯标准库,不依赖 PIL),供 VLM 带图请求使用。
- orange: 纯橙色。模型若真"看到"图,问颜色应答橙色;瞎猜则说明视觉链路没通。
- square: 白底黑方块。用于形状/内容识别的第二个判别样本。
输出 test_images.json,内容为 OpenAI image_url 用的 data URI。
"""
import base64
import json
import os
import struct
import zlib
def png_chunk(tag, data):
return (
struct.pack(">I", len(data))
+ tag
+ data
+ struct.pack(">I", zlib.crc32(tag + data) & 0xFFFFFFFF)
)
def write_png(w, h, pixel):
raw = bytearray()
for y in range(h):
raw.append(0) # filter type 0 (None)
for x in range(w):
raw.extend(pixel(x, y))
comp = zlib.compress(bytes(raw), 9)
return (
b"\x89PNG\r\n\x1a\n"
+ png_chunk(b"IHDR", struct.pack(">IIBBBBB", w, h, 8, 2, 0, 0, 0))
+ png_chunk(b"IDAT", comp)
+ png_chunk(b"IEND", b"")
)
def main():
# 448x448:与 vision 模型输入尺寸一致,避免 server 额外缩放。
w = h = 448
orange = write_png(w, h, lambda x, y: (255, 128, 0))
def square_px(x, y):
if 144 <= x < 304 and 144 <= y < 304:
return (0, 0, 0)
return (255, 255, 255)
square = write_png(w, h, square_px)
out = {
"orange": "data:image/png;base64," + base64.b64encode(orange).decode(),
"square": "data:image/png;base64," + base64.b64encode(square).decode(),
}
path = os.path.join(os.path.dirname(os.path.abspath(__file__)), "test_images.json")
with open(path, "w") as f:
json.dump(out, f)
print("orange png bytes:", len(orange), "square png bytes:", len(square))
print("orange data uri len:", len(out["orange"]))
print("square data uri len:", len(out["square"]))
print("saved:", path)
if __name__ == "__main__":
main()