Enhancing interpolated thermal images

1. A three-dimensional printer comprising:

a first camera to capture a low-resolution thermal image of a build material bed;

an interpolation engine to generate an interpolated thermal image based on the low-resolution thermal image; and

a correction engine to enhance fine details and remove undesirable gradients and interpolation artifacts of the interpolated thermal image without distorting thermal values from portions of the interpolated thermal image without fine details to produce an enhanced thermal image, wherein the correction engine enhances the fine details and removes undesirable gradients and interpolation artifacts using a machine learning model trained to enhance the fine details without distorting the other portions of the interpolated thermal image, and wherein the machine learning model is trained using a high-resolution thermal image from a second camera.

2. The three-dimensional printer of claim 1 , further comprising a training engine to update the machine learning model based on the low-resolution thermal image to train the machine learning model to avoid distorting the thermal values and based on a high-resolution thermal image from the second camera to train the machine learning model to enhance the fine details.

3. The three-dimensional printer of claim 1 , further comprising a prediction engine to compare the enhanced thermal image to a predicted thermal image, and a print process engine to adjust a print parameter based on the comparison of the enhanced thermal image to the predicted thermal image.

4. The three-dimensional printer of claim 1 , wherein the interpolation engine is to receive a shape image and generate the interpolated thermal image based on the low-resolution thermal image and the shape image.

5. A method, comprising:

generating an interpolated thermal image based on a low-resolution thermal image from a first camera;

enhancing the interpolated thermal image using a machine learning model to produce an enhanced thermal image, wherein enhancing the interpolated thermal image includes enhancing fine details of the interpolated thermal image and removing undesirable gradients and interpolation artifacts;

calculating a loss function based on the low-resolution thermal image from the first camera and a high-resolution thermal image from a second camera; and

updating the machine learning model based on the loss function.

6. The method of claim 5 , wherein calculating the loss function comprises calculating the loss function based on a first comparison of thermal values of the enhanced thermal image to thermal values of the interpolated thermal image from the first camera and a second comparison of gradients of the enhanced thermal image to gradients of the high-resolution thermal image from the second camera.

7. The method of claim 6 , further comprising applying a mask to the enhanced thermal image, applying a mask to the interpolated thermal image, and comparing the masked enhanced thermal image to the masked interpolated thermal image to produce the first comparison.

8. The method of claim 7 , further comprising generating the mask based on a shape image indicating locations of a build material bed to be fused for a layer of the build material bed corresponding to the low-resolution thermal image.

9. The method of claim 6 , wherein calculating the loss function comprises applying a first weight to the first comparison and a second weight to the second comparison and calculating the loss function based on the first comparison weighted with the first weight and the second comparison weighted by the second weight.

10. A non-transitory computer-readable medium comprising instructions that, when executed by a processor, cause the processor to:

generate an interpolated thermal image based on a low-resolution thermal image of a build material bed; and

enhance fine details of the interpolated thermal image using a machine learning model to produce an enhanced thermal image, the machine learning model designed to enhance the fine details and remove undesirable gradients and interpolation artifacts while maintaining accuracy of thermal values from part or powder centers, wherein the machine learning model is trained using a high-resolution thermal image.

11. The computer-readable medium of claim 10 , further comprising instructions that cause the processor to cause a printer to form a first layer of a three-dimensional object on the build material bed, cause the printer to capture the low-resolution thermal image, adjust a print parameter based on the enhanced thermal image, and cause the printer to form a second layer of the three-dimensional object on the build material bed according to the adjusted print parameter.

12. The computer-readable medium of claim 10 , wherein the machine learning model comprises a neural network including a layer with a plurality of kernel sizes.

13. The computer-readable medium of claim 10 , further comprising instructions that cause the processor to generate a predicted thermal image based on a shape image, wherein the instructions to generate the interpolated thermal image include instructions that cause the processor to generate the interpolated thermal image based on the predicted thermal image.

14. The computer-readable medium of claim 13 , wherein the instructions to generate the interpolated thermal image include instructions that cause the processor to downsample the predicted thermal image to produce a downsampled predicted thermal image, calculate a ratio image based on a ratio of the downsampled predicted thermal image to the low-resolution thermal image, interpolate the ratio image to produce an interpolated ratio image, and multiply the predicted thermal image by the interpolated ratio image to produce the interpolated thermal image.