Train movinet using hmdb51, and show it's predictions in classifications. #6
Conversation
…e split, and specifying the code.
This lets DL work on mac OS. DL is not the bottleneck anyways, it's the train loop.
|
The problem on Mac was multiprocessing in pytorch dataloader. DL with num_workers lets the DL to iterate, as it runs in the main process. |
| """ | ||
| Dataset used to parallelize the reading of the timestamps | ||
| of a list of videos, given their paths in the filesystem. | ||
|
|
There was a problem hiding this comment.
The comment is not explaining much. but you did say you plan to add more comments. so that's good. if we need a class hierarchy diagram, maybe that helps
There was a problem hiding this comment.
Oh actually, this is the exact doc string of the pytorch version which we have overriden. Let me change it to why this is needed instead.
| "id": "c9b8f6e9-668c-4c4b-81dd-5fcfb7956d51", | ||
| "metadata": {}, | ||
| "source": [ | ||
| "# Classification with a vanilla model vs trained model.\n", |
There was a problem hiding this comment.
| "# Classification with a vanilla model vs trained model.\n", | |
| "# Activity Recognition on Videos Stored in ApertureDB\n", |
There was a problem hiding this comment.
I am debating if it should address finetuninig in the title but I dont suppose we show trianing example, do we?
| "def get_common():\n", | ||
| " \"\"\"\n", | ||
| " Just common parameters.\n", | ||
| " Applies to the training and data loading sections.\n", |
There was a problem hiding this comment.
Code does need more comments but its looking good. Are you planning to land the READMe in docs?
| "id": "4f7045eb-228a-4019-93a1-7fe69ee1068a", | ||
| "metadata": {}, | ||
| "source": [ | ||
| "## Make a model from trained movinet with hmdb51" |
| "id": "a7c4bda9", | ||
| "metadata": {}, | ||
| "source": [ | ||
| "## See the shape of the tensor passsed through model.\n", |
There was a problem hiding this comment.
I guess I Don't follow what seeing the shape means
| @@ -0,0 +1,41 @@ | |||
| # Using HMDB51 to train movinet. | |||
|
|
|||
There was a problem hiding this comment.
Should this have more context ?
| import os | ||
| import subprocess | ||
|
|
||
| from aperturedb.ParallelLoader import ParallelLoader |
There was a problem hiding this comment.
what is this file? as in, needs comments or mention in readme
Use HMDB51 to train movinet. While this is an end to end example that works, it needs the following addressed to make it to the docs.