search

ArtyZe / yolov3_lite

yolov3 model compress and acceleration (quantization, sparse)， c++ version
36 stars 9 forks source link

Predictions are not drawing after testing trained model #4

Open ashank-art opened 5 years ago

ashank-art commented 5 years ago

@ArtyZe Predictions are not drawing on the image after testing a newly trained model, is there any file that will store the predictions??

ArtyZe commented 5 years ago

@ashank-art Natuerlich :) It's same with the normal Yolo, you can see the result in prediction.png

ashank-art commented 5 years ago

@ArtyZe ..thanks for your prompt reply! The prediction.png image does not have any bounding box predictions. It is the same as the original image.

ArtyZe commented 5 years ago

@ashank-art can you show me your train log? what is the loss value when you finish your train process?

abhigoku10 commented 5 years ago

@ArtyZe even i am facing he same issue. Just wanted to know should we train the custom model from scratch with the darknet weights or unpruned weights

ArtyZe commented 5 years ago

Both is ok :) first can reach the final result faster, but result is all same

On 07/11/2019 19:44, abhigoku10 wrote:

@ArtyZe even i am facing he same issue. Just wanted to know should we train the custom model from scratch with the darknet weights or unpruned weights

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

abhigoku10 commented 5 years ago

@ArtyZe the quantization applied is only for CPU or we could be using for GPU platform also , has the model size reduced ?

ArtyZe commented 5 years ago

as for I want to use it to cpu，but hash net and low rank convolutional and quantization is also fit for gpu，both are ok. But you need to know，it’s not easy to change source code of cuda

On 07/12/2019 16:44, abhigoku10 wrote:

@ArtyZe the quantization applied is only for CPU or we could be using for GPU platform also , has the model size reduced ?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

abhigoku10 commented 5 years ago

@ArtyZe thanks for instant response as always . I am training your code base with custom data, i shall just have a look into it if am able to run it on GPU, the intention is wanted to run the model with high fps in gpu also