Yolov5 in EdgeRealtimeVideoAnalytics: issue importing pytorch model

[davave]

Dear all,

First of all many thanks for this project.
I am new both to Redis and deep learning so forgive me if i'm making trivial mistakes.

I am trying to run YoloV5 model within EdgeRealtimeVideoAnalytics framework, but i am having troubles. that's what I did:

1. first I exported the model with torchscript using this script provided by Ultralytics

   # TorchScript export ----------------------------------------------------------------------------------------------- 
    prefix = colorstr('TorchScript:') 
    try:  
        print(f'\n{prefix} starting export with torch {torch.__version__}...')  
        f = opt.weights.replace('.pt', '.torchscript.pt')  # filename  
        ts = torch.jit.trace(model, img, strict=False)  
        (optimize_for_mobile(ts) if opt.optimize else ts).save(f)  
        print(f'{prefix} export success, saved as {f} ({file_size(f):.1f} MB)')  
    except Exception as e:  
        print(f'{prefix} export failure: {e}')

I got the file 23weights.torchscript.pt

2. I then modified init.py (obviously putting the .pt file in the right directory)

    with open('models/23weights.torchscript.pt', 'rb') as f:
        model = f.read()
        #res = conn.execute_command('AI.MODELSET', 'yolo:model', 'TF', args.device, 'INPUTS', 'input', 'OUTPUTS', 'output', model)
        res = conn.execute_command('AI.MODELSET', 'yolo:model', 'TORCH', args.device, model)
        print(res)

but i get the following message:

init_1                                | Loading model - Traceback (most recent call last):
init_1                                |   File "init.py", line 37, in <module>
init_1                                |     res = conn.execute_command('AI.MODELSET', 'yolo:model', 'TORCH', args.device, model)
init_1                                |   File "/usr/local/lib/python3.7/site-packages/redis/client.py", line 775, in execute_command
init_1                                |     return self.parse_response(connection, command_name, **options)
init_1                                |   File "/usr/local/lib/python3.7/site-packages/redis/client.py", line 789, in parse_response
init_1                                |     response = connection.read_response()
init_1                                |   File "/usr/local/lib/python3.7/site-packages/redis/connection.py", line 642, in read_response
init_1                                |     raise response
init_1                                | redis.exceptions.ResponseError: version_number <= kMaxSupportedFileFormatVersion INTERNAL ASSERT FAILED at /pytorch/caffe2/serialize/inline_container.cc:131, please report a bug to PyTorch. Attempted to read a PyTorch file with version 3, but the maximum supported version for reading is 1. Your PyTorch installation may be too old. (init at /pytorch/caffe2/serialize/inline_container.cc:131) frame #0: c10::Error::Error(c10::SourceLocation, std::string const&) + 0x33 (0x7f12e31a6273 in /usr/lib/redis/modules/backends/redisai_torch/lib/libc10.so) frame #1: caffe2::serialize::PyTorchStreamReader::init() + 0x1e9a (0x7f12e500040a in /usr/lib/redis/modules/backends/redisai_torch/lib/libtorch.so) frame #2: caffe2::serialize::PyTorchStreamReader::PyTorchStreamReader(std::unique_ptr<caffe2::serialize::ReadAdapterInterface, std::default_delete<caffe2::serialize::ReadAdapterInterface> >) + 0x53 (0x7f12e50015c3 in /usr/lib/redis/modules/backends/redisai_torch/lib/libtorch.so) frame #3: <unknown function> + 0x2d120d2 (0x7f12e60df0d2 in /usr/lib/redis/modules/backends/redisai_torch/lib/libtorch.so) frame #4: torch::jit::load(std::unique_ptr<caffe2::serialize::ReadAdapterInterface, std::default_delete<caffe2::serialize::ReadAdapterInterface> >, c10::optional<c10::Device>, std::unordered_map<std::string, std::string, std::hash<std::string>, std::equal_to<std::string>, std::allocator<std::pair<std::string const, std::string> > >&) + 0x27 (0x7f12e60de027 in /usr/lib/redis/modules/backends/redisai_torch/lib/libtorch.so) frame #5: torch::jit::load(std::istream&, c10::optional<c10::Device>, std::unordered_map<std::string, std::string, std::hash<std::string>, std::equal_to<std::string>, std::allocator<std::pair<std::string const, std::string> > >&) + 0x69 (0x7f12e60de2a9 in /usr/lib/redis/modules/backends/redisai_torch/lib/libtorch.so) frame #6: torchLoadModel + 0x1f7 (0x7f12ef5e48b7 in /usr/lib/redis/modules/backends/redisai_torch/redisai_torch.so) frame #7: RAI_ModelCreateTorch + 0x9d (0x7f12ef5e0ebd in /usr/lib/redis/modules/backends/redisai_torch/redisai_torch.so) frame #8: RedisAI_ModelSet_RedisCommand + 0x242 (0x7f12fa57b782 in /usr/lib/redis/modules/redisai.so) frame #9: RedisModuleCommandDispatcher + 0x56 (0x564747ba66d6 in redis-server *:6379) frame #10: call + 0xa7 (0x564747b362b7 in redis-server *:6379) frame #11: processCommand + 0x51e (0x564747b36b5e in redis-server *:6379) frame #12: processInputBuffer + 0x171 (0x564747b46d51 in redis-server *:6379) frame #13: aeProcessEvents + 0x101 (0x564747b302e1 in redis-server *:6379) frame #14: aeMain + 0x2b (0x564747b306eb in redis-server *:6379) frame #15: main + 0x4b9 (0x564747b2d569 in redis-server *:6379) frame #16: __libc_start_main + 0xeb (0x7f12fa5b809b in /lib/x86_64-linux-gnu/libc.so.6) frame #17: _start + 0x2a (0x564747b2d7aa in redis-server *:6379) 
edgerealtimevideoanalytics_init_1 exited with code 1

As you can see the problem is related with pytorch version, but I can't figure out how to fix this problem. Could you please help me? I am not even sure I am passing to RedisAI the right formatted model, so please let me know if I am making mistakes.

Thank you
Davide