Update README.md
Browse files
README.md
CHANGED
@@ -15,7 +15,7 @@ This version was exported using the old torch to ONNX converter (`torch.onnx.exp
|
|
15 |
|
16 |
**Notes:**
|
17 |
1. **Custom FourierUnitJIT**: A custom [FourierUnitJIT](https://github.com/Carve-Photos/lama/blob/main/saicinpainting/training/modules/ffc.py) implementation is used since the original cannot be directly ported to ONNX without overhead. The result is identical to the original model.
|
18 |
-
2. **Fixed Input Shape**: The input shape is fixed at
|
19 |
3. **Opset Version 17**: This model uses opset version 17.
|
20 |
|
21 |
### 2. `lama.onnx` (NOT RECOMMENDED)
|
@@ -38,7 +38,11 @@ This version was exported using the new torch to ONNX converter (`torch.onnx.dyn
|
|
38 |
**Original image:**
|
39 |
![original image](./image.jpg)
|
40 |
|
41 |
-
|
|
|
|
|
|
|
|
|
42 |
![onnx output](./output_onnx.png)
|
43 |
|
44 |
**Original model output:**
|
|
|
15 |
|
16 |
**Notes:**
|
17 |
1. **Custom FourierUnitJIT**: A custom [FourierUnitJIT](https://github.com/Carve-Photos/lama/blob/main/saicinpainting/training/modules/ffc.py) implementation is used since the original cannot be directly ported to ONNX without overhead. The result is identical to the original model.
|
18 |
+
2. **Fixed Input Shape**: The input shape is fixed at 800x800 pixels. Although dynamic input shapes are possible, they would require resolving issues with dynamic padding in the `irfft` and `rfftn` functions in `ffc.py`.
|
19 |
3. **Opset Version 17**: This model uses opset version 17.
|
20 |
|
21 |
### 2. `lama.onnx` (NOT RECOMMENDED)
|
|
|
38 |
**Original image:**
|
39 |
![original image](./image.jpg)
|
40 |
|
41 |
+
|
42 |
+
**lama_fp32.onnx - output:**
|
43 |
+
![onnx output](./output_onnx_fp32.png)
|
44 |
+
|
45 |
+
**lama.onnx - output:**
|
46 |
![onnx output](./output_onnx.png)
|
47 |
|
48 |
**Original model output:**
|