Remove dead code comment

tracel-ai · Mar 19, 2024 · 7bff1e5 · 7bff1e5
1 parent 1cb8f5c
commit 7bff1e5
Showing 1 changed file with 0 additions and 107 deletions.
diff --git a/yolox-burn/src/model/boxes.rs b/yolox-burn/src/model/boxes.rs
@@ -139,110 +139,3 @@ pub fn non_maximum_suppression(bboxes: &mut [Vec<BoundingBox>], threshold: f32)
         bboxes_for_class.truncate(current_index);
     }
 }
-
-// pub fn nms<B: Backend>(
-//     boxes: Tensor<B, 3>,
-//     scores: Tensor<B, 3>,
-//     iou_threshold: f64,
-//     score_threshold: f64,
-// ) -> () {
-//     // Tensor<B, 2>
-//     let [_, num_boxes, num_classes] = scores.shape().dims;
-//     let device = boxes.device();
-//     let is_valid = Tensor::<B, 1, Bool>::from_data([true; 1], &device);
-//     // let values: Tensor<B, 1, _> = Tensor::from_data([1, 0, 3, 4, -1], &device);
-
-//     let x = Tensor::arange(0..6 as i64, &device).reshape(Shape::new([2, 3]));
-//     let mask = x.greater_elem(1);
-
-//     let selected_indices = boxes
-//         .iter_dim(0)
-//         .enumerate()
-//         // Per-batch NMS
-//         .map(|(batch_idx, candidate_boxes)| {
-//             // Per-class filtering
-//             (0..num_classes).map(|cls_idx| {
-//                 scores.select(
-//                     0,
-//                     Tensor::<B, 1, Int>::full(Shape::new([1]), batch_idx as i64, &device),
-//                 );
-//                 // [batch_size, num_boxes, 1]
-//                 let scores_mask = scores.greater_equal_elem(score_threshold).any_dim(2);
-//                 // 1. `non_zero` to get the mask's valid boxes indices
-//                 // 2. `select` to return only valid boxes (by score threshold)
-//                 // 3. `argsort` to get the indices ranked by classification score
-//                 // NOTE: we could do it all on CPU (preds to_vec) to iterate and compare as done in
-//                 // [candle](https://github.com/huggingface/candle/blob/main/candle-examples/examples/yolo-v3/main.rs#L38)
-//                 // ([nms](https://github.com/huggingface/candle/blob/main/candle-transformers/src/object_detection.rs#L31))
-//                 // but this is not so nice
-//                 let box_idx = scores_mask
-//                     .iter_dim(1)
-//                     .filter_map(|box_idx| {
-//                         // NOTE: bool tensors do not support `select` and `into_scalar()`
-//                         if scores_mask
-//                             .slice([batch_idx..batch_idx + 1])
-//                             .any()
-//                             .int()
-//                             .into_scalar() as bool
-//                         {
-//                             Some(box_idx)
-//                         } else {
-//                             None
-//                         }
-//                     })
-//                     .collect();
-//                 // scores[batch_idx, ]
-//                 // [batch_size, num_boxes, 1]
-//                 let scores_max = scores.gather(dim, indices);
-//             })
-//         });
-// }
-
-// std::vector<SelectedIndex> selected_indices;
-//   std::vector<BoxInfoPtr> selected_boxes_inside_class;
-//   selected_boxes_inside_class.reserve(std::min<size_t>(static_cast<size_t>(max_output_boxes_per_class), pc.num_boxes_));
-
-//   for (int64_t batch_index = 0; batch_index < pc.num_batches_; ++batch_index) {
-//     for (int64_t class_index = 0; class_index < pc.num_classes_; ++class_index) {
-//       int64_t box_score_offset = (batch_index * pc.num_classes_ + class_index) * pc.num_boxes_;
-//       const float* batch_boxes = boxes_data + (batch_index * pc.num_boxes_ * 4);
-//       std::vector<BoxInfoPtr> candidate_boxes;
-//       candidate_boxes.reserve(pc.num_boxes_);
-
-//       // Filter by score_threshold_
-//       const auto* class_scores = scores_data + box_score_offset;
-//       if (pc.score_threshold_ != nullptr) {
-//         for (int64_t box_index = 0; box_index < pc.num_boxes_; ++box_index, ++class_scores) {
-//           if (*class_scores > score_threshold) {
-//             candidate_boxes.emplace_back(*class_scores, box_index);
-//           }
-//         }
-//       } else {
-//         for (int64_t box_index = 0; box_index < pc.num_boxes_; ++box_index, ++class_scores) {
-//           candidate_boxes.emplace_back(*class_scores, box_index);
-//         }
-//       }
-//       std::priority_queue<BoxInfoPtr, std::vector<BoxInfoPtr>> sorted_boxes(std::less<BoxInfoPtr>(), std::move(candidate_boxes));
-
-//       selected_boxes_inside_class.clear();
-//       // Get the next box with top score, filter by iou_threshold
-//       while (!sorted_boxes.empty() && static_cast<int64_t>(selected_boxes_inside_class.size()) < max_output_boxes_per_class) {
-//         const BoxInfoPtr& next_top_score = sorted_boxes.top();
-
-//         bool selected = true;
-//         // Check with existing selected boxes for this class, suppress if exceed the IOU (Intersection Over Union) threshold
-//         for (const auto& selected_index : selected_boxes_inside_class) {
-//           if (SuppressByIOU(batch_boxes, next_top_score.index_, selected_index.index_, center_point_box, iou_threshold)) {
-//             selected = false;
-//             break;
-//           }
-//         }
-
-//         if (selected) {
-//           selected_boxes_inside_class.push_back(next_top_score);
-//           selected_indices.emplace_back(batch_index, class_index, next_top_score.index_);
-//         }
-//         sorted_boxes.pop();
-//       }  // while
-//     }    // for class_index
-//   }      // for batch_index