Honored ContributorI have designed an OpenCL-based Accelerator for Convolutional Neural Networks on FPGAs, it can be found on github.com/doonny/PipeCNN.
Currently, AlexNet and VGG-166 models are tested on DE5-net boards. Hope someone maybe interested to join us to improve the design. Cheers,
Honored ContributorI'm very interested in your work，and thank you for your code, but i can't compile the conv_pipe.cl because I can't find the RTL folder.
Honored ContributorHonored ContributorHi,
Thanks for providing code. I am able to run the demo using OpenCV with following o/p for Alexnet: Loading kernel/binary from file project/conv.aocx Reprogramming device [0] with handle 1 61063552 total weights read 1024 total output reference read Loading picture ./data/ILSVRC2012_val_00000001.JPEG ..... Executing Layer 1: Launching single work-item kernel winbuffer Launching single work-item kernel Conv Launching single work-item kernel Pooling Launching kernel MemWr with local size: 1, 1, 16 (global size: 27, 27, 96) Launching kernel lrn with local size: 1, 1, 24 (global size: 27, 27, 24) Executing Layer 2: Launching single work-item kernel winbuffer Launching single work-item kernel Conv Launching single work-item kernel Pooling Launching kernel MemWr with local size: 1, 1, 16 (global size: 13, 13, 256) Launching kernel lrn with local size: 1, 1, 64 (global size: 13, 13, 64) Executing Layer 3: Launching single work-item kernel winbuffer Launching single work-item kernel Conv Launching kernel MemWr with local size: 1, 1, 16 (global size: 13, 13, 384) Executing Layer 4: Launching single work-item kernel winbuffer Launching single work-item kernel Conv Launching kernel MemWr with local size: 1, 1, 16 (global size: 13, 13, 384) Executing Layer 5: Launching single work-item kernel winbuffer Launching single work-item kernel Conv Launching single work-item kernel Pooling Launching kernel MemWr with local size: 1, 1, 16 (global size: 6, 6, 256) Executing Layer 6: Launching single work-item kernel winbuffer Launching single work-item kernel Conv Launching kernel MemWr with local size: 1, 1, 16 (global size: 1, 1, 4096) Executing Layer 7: Launching single work-item kernel winbuffer Launching single work-item kernel Conv Launching kernel MemWr with local size: 1, 1, 16 (global size: 1, 1, 4096) Executing Layer 8: Launching single work-item kernel winbuffer Launching single work-item kernel Conv Launching kernel MemWr with local size: 1, 1, 16 (global size: 1, 1, 1024) Done! Inference time is 0.045396s Copied all batched results from fc_2 buffers. Selected item = 0 from the combined batch results in fc buffers The inference result is n01737021 water snake (the prob is 37.00) False: True_label = 970 Inferred_label = 58 Current Top-1 accuracy = 0.000 Current Top-5 accuracy = 1.000 Loading picture ./data/ILSVRC2012_val_00000002.JPEG ..... Done! Inference time is 0.044439s Copied all batched results from fc_2 buffers. Selected item = 0 from the combined batch results in fc buffers The inference result is n04228054 ski (the prob is 46.00) False: True_label = 230 Inferred_label = 795 Current Top-1 accuracy = 0.000 Current Top-5 accuracy = 1.000 Loading picture ./data/ILSVRC2012_val_00000003.JPEG ..... Done! Inference time is 0.044546s Copied all batched results from fc_2 buffers. Selected item = 0 from the combined batch results in fc buffers The inference result is n02105855 Shetland sheepdog, Shetland sheep dog, Shetland (the prob is 99.00) Current Top-1 accuracy = 0.333 Current Top-5 accuracy = 1.000 Loading picture ./data/ILSVRC2012_val_00000004.JPEG ..... Done! Inference time is 0.044284s Copied all batched results from fc_2 buffers. Selected item = 0 from the combined batch results in fc buffers The inference result is n07920052 espresso (the prob is 81.00) False: True_label = 516 Inferred_label = 967 Current Top-1 accuracy = 0.250 Current Top-5 accuracy = 0.750 Loading picture ./data/ILSVRC2012_val_00000005.JPEG ..... Done! Inference time is 0.044279s Copied all batched results from fc_2 buffers. Selected item = 0 from the combined batch results in fc buffers The inference result is n03125729 cradle (the prob is 34.00) Current Top-1 accuracy = 0.400 Current Top-5 accuracy = 0.800 Loading picture ./data/ILSVRC2012_val_00000006.JPEG ..... Done! Inference time is 0.044487s Copied all batched results from fc_2 buffers. Selected item = 0 from the combined batch results in fc buffers The inference result is n01755581 diamondback, diamondback rattlesnake, Crotalus adamanteus (the prob is 51.00) False: True_label = 334 Inferred_label = 67 Current Top-1 accuracy = 0.333 Current Top-5 accuracy = 0.833 Loading picture ./data/ILSVRC2012_val_00000007.JPEG ..... Done! Inference time is 0.044391s Copied all batched results from fc_2 buffers. Selected item = 0 from the combined batch results in fc buffers The inference result is n02346627 porcupine, hedgehog (the prob is 99.00) Current Top-1 accuracy = 0.429 Current Top-5 accuracy = 0.857 Loading picture ./data/ILSVRC2012_val_00000008.JPEG ..... Done! Inference time is 0.044590s Copied all batched results from fc_2 buffers. Selected item = 0 from the combined batch results in fc buffers The inference result is n03063599 coffee mug (the prob is 56.00) False: True_label = 674 Inferred_label = 504 Current Top-1 accuracy = 0.375 Current Top-5 accuracy = 0.750 Loading picture ./data/ILSVRC2012_val_00000009.JPEG ..... Done! Inference time is 0.044468s Copied all batched results from fc_2 buffers. Selected item = 0 from the combined batch results in fc buffers The inference result is n03201208 dining table, board (the prob is 26.00) False: True_label = 332 Inferred_label = 532 Current Top-1 accuracy = 0.333 Current Top-5 accuracy = 0.667 Loading picture ./data/ILSVRC2012_val_00000010.JPEG ..... Done! Inference time is 0.044281s Copied all batched results from fc_2 buffers. Selected item = 0 from the combined batch results in fc buffers The inference result is n04399382 teddy, teddy bear (the prob is 36.00) False: True_label = 0 Inferred_label = 850 Current Top-1 accuracy = 0.300 Current Top-5 accuracy = 0.700 Demo exited !!! Total number of 10 pictures have been processed. Final Top-1 accuracy = 0.300 Final Top-5 accuracy = 0.700 Any suggestion to improve the results using any optimization techniques...? Thanks
