Improve YOLO test speed #603
Conversation
|
@Ben-Epstein removed the print statements. essentially the model logger was failing if there were no predictions for the batch at all. therefore we need to fallback to empty arrays and make sure the shapes are equal. remove the print statements |
Codecov Report
@@ Coverage Diff @@
## main #603 +/- ##
==========================================
- Coverage 84.21% 84.02% -0.19%
==========================================
Files 165 165
Lines 13005 13036 +31
==========================================
+ Hits 10952 10954 +2
- Misses 2053 2082 +29
... and 2 files with indirect coverage changes 📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more |
Co-authored-by: Ben Epstein <[email protected]>
| prob = np.zeros(gold_prob_shape) | ||
| if len(pred_prob_arrays) > 0: | ||
| prob = np.concatenate([pred_prob_arrays, np.zeros(gold_prob_shape)]) | ||
| # prob = np.concatenate([pred_prob_arrays, np.zeros(gold_prob_shape)]) |
There was a problem hiding this comment.
want to cleanup the older np.concat ?
There was a problem hiding this comment.
Tested with our publicly facing notebook and a YAML file without the key path https://console.dev.rungalileo.io/insights?projectId=543aedcf-2eb5-422c-95f0-c2e641ef15da&runId=7dc85dce-c64d-480c-90df-8546c7623c55&insightsTab=metrics&taskType=5. Looks good to me!
| prob = np.zeros(gold_prob_shape) | ||
| if len(pred_prob_arrays) > 0: | ||
| prob = np.concatenate([pred_prob_arrays, np.zeros(gold_prob_shape)]) | ||
| # prob = np.concatenate([pred_prob_arrays, np.zeros(gold_prob_shape)]) |
There was a problem hiding this comment.
| # prob = np.concatenate([pred_prob_arrays, np.zeros(gold_prob_shape)]) | |
| # prob = np.concatenate([pred_prob_arrays, np.zeros(gold_prob_shape)]) |
Instead of training on coco128 we create a custom dataset and we only have 4 samples