gray model

[chenchunhu]

I used the model you provided and tried to convert it into an ncnn model, but an error occurred.

[linyicheng1]

Try use the model in pytorch first, and then convert it to ncnn.

[zhixuan4123]

I encountered the same problem. Could you please provide detailed instructions on how to convert it? Or offer relevant conversion code? Thank you!

[deepConnectionism]

I had the same problem:

I used the following command to convert to the NCNN model:

./onnx2ncnn /root/LET-NET/model/letnet-gray.onnx /root/LET-NET/model/letnet-gray.param /root/LET-NET/model/letnet-gray.bin
./ncnnoptimize /root/LET-NET/model/letnet-gray.param /root/LET-NET/model/letnet-gray.bin /root/LET-NET/model/letnet-gray-opt.param /root/LET-NET/model/letnet-gray-opt.bin 0

when I run vins-mono, meet：

find_blob_index_by_name score failed
Try
    ex.extract("27", out0);
    ex.extract("33", out1);
find_blob_index_by_name descriptor failed
Try
    ex.extract("27", out0);
    ex.extract("33", out1);
Segmentation fault (core dumped)

How Do I Fix This? Can you provide the conversion code?

[deepConnectionism]

I solved the problem:

import onnx

# 加载 ONNX 模型
model = onnx.load("letnet-gray.onnx")

# 获取图
graph = model.graph

# 修改输入节点名称
for input in graph.input:
    if input.name == 'input.1':
        input.name = 'input'

# 修改输出节点名称
for output in graph.output:
    if output.name == '33':
        output.name = 'descriptor'
    elif output.name == '27':
        output.name = 'score'

# 修改节点的输出名称
for node in graph.node:
    for i, input_name in enumerate(node.input):
        if input_name == 'input.1':
            node.input[i] = 'input'

    for i, output in enumerate(node.output):
        if output == '33':
            node.output[i] = 'descriptor'
        elif output == '27':
            node.output[i] = 'score'

# 保存修改后的模型
onnx.save(model, "letnet-gray-modify.onnx")

Then convert the model from onnx to NCNN.

But I found that feature tracking didn't get any better.